diff --git a/README b/README
index a85962e..94a1b40 100644
--- a/README
+++ b/README
@@ -50,6 +50,7 @@
 	-l Generate per-job latency logs
 	-w Generate per-job bandwidth logs
 	-f <file> Read <file> for job descriptions
+	-O <file> Log output to file
 	-h Print help info
 	-v Print version information and exit
 
diff --git a/fio.c b/fio.c
index 93ca720..c95d7b5 100644
--- a/fio.c
+++ b/fio.c
@@ -222,7 +222,7 @@
 
 		rate = (td->this_io_bytes[ddir] - td->rate_bytes) / spent;
 		if (rate < td->ratemin) {
-			printf("%s: min rate %d not met, got %ldKiB/sec\n", td->name, td->ratemin, rate);
+			fprintf(f_out, "%s: min rate %d not met, got %ldKiB/sec\n", td->name, td->ratemin, rate);
 			if (rate_quit)
 				terminate_threads(td->groupid);
 			return 1;
@@ -274,8 +274,8 @@
 	int i;
 
 	for (i = 0; i < len; i++)
-		printf("%02x", p[i]);
-	printf("\n");
+		fprintf(f_out, "%02x", p[i]);
+	fprintf(f_out, "\n");
 }
 
 static int verify_io_u_crc32(struct verify_header *hdr, struct io_u *io_u)
@@ -287,8 +287,8 @@
 	c = crc32(p, hdr->len - sizeof(*hdr));
 
 	if (c != hdr->crc32) {
-		fprintf(stderr, "crc32: verify failed at %llu/%u\n", io_u->offset, io_u->buflen);
-		fprintf(stderr, "crc32: wanted %lx, got %lx\n", hdr->crc32, c);
+		fprintf(f_err, "crc32: verify failed at %llu/%u\n", io_u->offset, io_u->buflen);
+		fprintf(f_err, "crc32: wanted %lx, got %lx\n", hdr->crc32, c);
 		return 1;
 	}
 
@@ -305,7 +305,7 @@
 	md5_update(&md5_ctx, p, hdr->len - sizeof(*hdr));
 
 	if (memcmp(hdr->md5_digest, md5_ctx.hash, sizeof(md5_ctx.hash))) {
-		fprintf(stderr, "md5: verify failed at %llu/%u\n", io_u->offset, io_u->buflen);
+		fprintf(f_err, "md5: verify failed at %llu/%u\n", io_u->offset, io_u->buflen);
 		hexdump(hdr->md5_digest, sizeof(hdr->md5_digest));
 		hexdump(md5_ctx.hash, sizeof(md5_ctx.hash));
 		return 1;
@@ -327,7 +327,7 @@
 	else if (hdr->verify_type == VERIFY_CRC32)
 		ret = verify_io_u_crc32(hdr, io_u);
 	else {
-		fprintf(stderr, "Bad verify type %d\n", hdr->verify_type);
+		fprintf(f_err, "Bad verify type %d\n", hdr->verify_type);
 		ret = 1;
 	}
 
@@ -851,7 +851,7 @@
 	else if (td->io_engine == FIO_SPLICEIO)
 		return fio_spliceio_init(td);
 	else {
-		fprintf(stderr, "bad io_engine %d\n", td->io_engine);
+		fprintf(f_err, "bad io_engine %d\n", td->io_engine);
 		return 1;
 	}
 }
@@ -878,7 +878,7 @@
 	} else if (td->mem_type == MEM_MMAP)
 		munmap(td->orig_buffer, td->orig_buffer_size);
 	else
-		fprintf(stderr, "Bad memory type %d\n", td->mem_type);
+		fprintf(f_err, "Bad memory type %d\n", td->mem_type);
 
 	td->orig_buffer = NULL;
 }
@@ -952,17 +952,17 @@
 		return 0;
 
 	if (!size) {
-		fprintf(stderr, "Need size for create\n");
+		fprintf(f_err, "Need size for create\n");
 		td_verror(td, EINVAL);
 		return 1;
 	}
 
 	if (!extend) {
 		oflags = O_CREAT | O_TRUNC;
-		printf("%s: Laying out IO file (%LuMiB)\n", td->name, size >> 20);
+		fprintf(f_out, "%s: Laying out IO file (%LuMiB)\n", td->name, size >> 20);
 	} else {
 		oflags = O_APPEND;
-		printf("%s: Extending IO file (%Lu -> %LuMiB)\n", td->name, (td->file_size - size) >> 20, td->file_size >> 20);
+		fprintf(f_out, "%s: Extending IO file (%Lu -> %LuMiB)\n", td->name, (td->file_size - size) >> 20, td->file_size >> 20);
 	}
 
 	td->fd = open(td->file_name, O_WRONLY | oflags, 0644);
@@ -1068,13 +1068,13 @@
 		return ret;
 
 	if (td->file_offset > td->real_file_size) {
-		fprintf(stderr, "%s: offset extends end (%Lu > %Lu)\n", td->name, td->file_offset, td->real_file_size);
+		fprintf(f_err, "%s: offset extends end (%Lu > %Lu)\n", td->name, td->file_offset, td->real_file_size);
 		return 1;
 	}
 
 	td->io_size = td->file_size;
 	if (td->io_size == 0) {
-		fprintf(stderr, "%s: no io blocks\n", td->name);
+		fprintf(f_err, "%s: no io blocks\n", td->name);
 		td_verror(td, EINVAL);
 		return 1;
 	}
@@ -1253,7 +1253,7 @@
 
 	sprintf(tmp2, "[%s]", td->ioscheduler);
 	if (!strstr(tmp, tmp2)) {
-		fprintf(stderr, "fio: io scheduler %s not found\n", td->ioscheduler);
+		fprintf(f_err, "fio: io scheduler %s not found\n", td->ioscheduler);
 		td_verror(td, EINVAL);
 		fclose(f);
 		return 1;
@@ -1461,7 +1461,7 @@
 			c = 'P';
 			break;
 		default:
-			printf("state %d\n", td->runstate);
+			fprintf(f_err, "state %d\n", td->runstate);
 	}
 
 	run_str[td->thread_number - 1] = c;
@@ -1613,7 +1613,7 @@
 	if (!nr_running && !nr_pending)
 		return;
 
-	printf("Threads now running (%d)", nr_running);
+	printf("Threads running: %d", nr_running);
 	if (m_rate || t_rate)
 		printf(", commitrate %d/%dKiB/sec", t_rate, m_rate);
 	if (eta_sec != INT_MAX) {
@@ -1678,8 +1678,7 @@
 	if (phys_mem) {
 		if ((mlock_size + 128 * 1024 * 1024) > phys_mem) {
 			mlock_size = phys_mem - 128 * 1024 * 1024;
-			printf("fio: limiting mlocked memory to %lluMiB\n",
-							mlock_size >> 20);
+			fprintf(f_out, "fio: limiting mlocked memory to %lluMiB\n", mlock_size >> 20);
 		}
 	}
 
@@ -1827,7 +1826,7 @@
 		}
 
 		if (left) {
-			fprintf(stderr, "fio: %d jobs failed to start\n", left);
+			fprintf(f_err, "fio: %d jobs failed to start\n", left);
 			for (i = 0; i < this_jobs; i++) {
 				td = map[i];
 				if (!td)
@@ -1877,7 +1876,7 @@
 		return 1;
 
 	if (!thread_number) {
-		printf("Nothing to do\n");
+		fprintf(f_err, "Nothing to do\n");
 		return 1;
 	}
 
diff --git a/fio.h b/fio.h
index 671cb22..f6c2ee3 100644
--- a/fio.h
+++ b/fio.h
@@ -324,6 +324,8 @@
 extern int thread_number;
 extern int shm_id;
 extern int groupid;
+extern FILE *f_out;
+extern FILE *f_err;
 
 extern struct thread_data *threads;
 
diff --git a/init.c b/init.c
index 3117d60..b40437a 100644
--- a/init.c
+++ b/init.c
@@ -54,6 +54,8 @@
 int write_bw_log = 0;
 int exitall_on_terminate = 0;
 unsigned long long mlock_size = 0;
+FILE *f_out = NULL;
+FILE *f_err = NULL;
 
 static struct thread_data *get_new_job(int global, struct thread_data *parent)
 {
@@ -89,13 +91,13 @@
 
 #ifndef FIO_HAVE_LIBAIO
 	if (td->io_engine == FIO_LIBAIO) {
-		fprintf(stderr, "Linux libaio not available\n");
+		fprintf(f_err, "Linux libaio not available\n");
 		return 1;
 	}
 #endif
 #ifndef FIO_HAVE_POSIXAIO
 	if (td->io_engine == FIO_POSIXAIO) {
-		fprintf(stderr, "posix aio not available\n");
+		fprintf(f_err, "posix aio not available\n");
 		return 1;
 	}
 #endif
@@ -172,9 +174,9 @@
 	ddir = td->ddir + (!td->sequential << 1) + (td->iomix << 2);
 
 	if (!job_add_num)
-		printf("%s: (g=%d): rw=%s, odir=%d, bs=%d-%d, rate=%d, ioengine=%s, iodepth=%d\n", td->name, td->groupid, ddir_str[ddir], td->odirect, td->min_bs, td->max_bs, td->rate, td->io_engine_name, td->iodepth);
+		fprintf(f_out, "%s: (g=%d): rw=%s, odir=%d, bs=%d-%d, rate=%d, ioengine=%s, iodepth=%d\n", td->name, td->groupid, ddir_str[ddir], td->odirect, td->min_bs, td->max_bs, td->rate, td->io_engine_name, td->iodepth);
 	else if (job_add_num == 1)
-		printf("...\n");
+		fprintf(f_out, "...\n");
 
 	/*
 	 * recurse add identical jobs, clear numjobs and stonewall options
@@ -493,7 +495,7 @@
 		return 0;
 	}
 
-	fprintf(stderr, "fio: data direction: read, write, randread, randwrite, rw, randrw\n");
+	fprintf(f_err, "fio: data direction: read, write, randread, randwrite, rw, randrw\n");
 	return 1;
 }
 
@@ -510,7 +512,7 @@
 		return 0;
 	}
 
-	fprintf(stderr, "fio: verify types: md5, crc32\n");
+	fprintf(f_err, "fio: verify types: md5, crc32\n");
 	return 1;
 }
 
@@ -527,7 +529,7 @@
 		return 0;
 	}
 
-	fprintf(stderr, "fio: mem type: malloc, shm, mmap\n");
+	fprintf(f_err, "fio: mem type: malloc, shm, mmap\n");
 	return 1;
 }
 
@@ -560,7 +562,7 @@
 		return 0;
 	}
 
-	fprintf(stderr, "fio: ioengine: { linuxaio, aio, libaio }, posixaio, sync, mmap, sgio, splice\n");
+	fprintf(f_err, "fio: ioengine: { linuxaio, aio, libaio }, posixaio, sync, mmap, sgio, splice\n");
 	return 1;
 }
 
@@ -618,7 +620,7 @@
 
 			if (!check_int(p, "prio", &prio)) {
 #ifndef FIO_HAVE_IOPRIO
-				fprintf(stderr, "io priorities not available\n");
+				fprintf(f_err, "io priorities not available\n");
 				ret = 1;
 				break;
 #endif
@@ -628,7 +630,7 @@
 			}
 			if (!check_int(p, "prioclass", &prioclass)) {
 #ifndef FIO_HAVE_IOPRIO
-				fprintf(stderr, "io priorities not available\n");
+				fprintf(f_err, "io priorities not available\n");
 				ret = 1;
 				break;
 #else
@@ -660,7 +662,7 @@
 			}
 			if (!check_int(p, "cpumask", &cpu)) {
 #ifndef FIO_HAVE_CPU_AFFINITY
-				fprintf(stderr, "cpu affinity not available\n");
+				fprintf(f_err, "cpu affinity not available\n");
 				ret = 1;
 				break;
 #endif
@@ -833,7 +835,7 @@
 			}
 			if (!check_strstore(p, "iolog", tmpbuf)) {
 				if (td->write_iolog) {
-					fprintf(stderr, "fio: read iolog overrides given write_iolog\n");
+					fprintf(f_err, "fio: read iolog overrides given write_iolog\n");
 					free(td->iolog_file);
 					td->write_iolog = 0;
 				}
@@ -847,7 +849,7 @@
 					td->iolog_file = strdup(tmpbuf);
 					td->write_iolog = 1;
 				} else
-					fprintf(stderr, "fio: read iolog overrides given write_iolog\n");
+					fprintf(f_err, "fio: read iolog overrides given write_iolog\n");
 				fgetpos(f, &off);
 				continue;
 			}
@@ -863,7 +865,7 @@
 			}
 			if (!check_strstore(p, "ioscheduler", tmpbuf)) {
 #ifndef FIO_HAVE_IOSCHED_SWITCH
-				fprintf(stderr, "io scheduler switching not available\n");
+				fprintf(f_err, "io scheduler switching not available\n");
 				ret = 1;
 				break;
 #else
@@ -961,7 +963,7 @@
 {
 	int c;
 
-	while ((c = getopt(argc, argv, "s:b:t:r:R:o:f:lwvh")) != EOF) {
+	while ((c = getopt(argc, argv, "s:b:t:r:R:o:f:lwvhO:")) != EOF) {
 		switch (c) {
 			case 's':
 				def_thread.sequential = !!atoi(optarg);
@@ -995,6 +997,14 @@
 			case 'w':
 				write_bw_log = 1;
 				break;
+			case 'O':
+				f_out = fopen(optarg, "w+");
+				if (!f_out) {
+					perror("fopen output");
+					exit(1);
+				}
+				f_err = f_out;
+				break;
 			case 'h':
 				usage(argv[0]);
 				exit(0);
@@ -1006,6 +1016,10 @@
 
 	if (!ini_file && argc > 1 && argv[argc - 1][0] != '-')
 		ini_file = strdup(argv[argc - 1]);
+	if (!f_out) {
+		f_out = stdout;
+		f_err = stderr;
+	}
 }
 
 static void free_shm(void)
@@ -1062,7 +1076,7 @@
 	parse_cmd_line(argc, argv);
 
 	if (!ini_file) {
-		printf("Need job file\n");
+		fprintf(f_err, "Need job file\n");
 		usage(argv[0]);
 		return 1;
 	}
diff --git a/ioengines.c b/ioengines.c
index b086b60..c94fcb3 100644
--- a/ioengines.c
+++ b/ioengines.c
@@ -610,7 +610,7 @@
 	int nr_blocks, lba;
 
 	if (io_u->buflen & (sd->bs - 1)) {
-		fprintf(stderr, "read/write not sector aligned\n");
+		fprintf(f_err, "read/write not sector aligned\n");
 		return EINVAL;
 	}
 
@@ -718,7 +718,7 @@
 		if (ret)
 			return ret;
 	} else {
-		fprintf(stderr, "ioengine sgio only works on block devices\n");
+		fprintf(f_err, "ioengine sgio only works on block devices\n");
 		return 1;
 	}
 
diff --git a/log.c b/log.c
index 8d313ec..a364e22 100644
--- a/log.c
+++ b/log.c
@@ -122,7 +122,7 @@
 		struct io_piece *ipo;
 
 		if (sscanf(p, "%d,%llu,%u", &rw, &offset, &bytes) != 3) {
-			fprintf(stderr, "bad iolog: %s\n", p);
+			fprintf(f_err, "bad iolog: %s\n", p);
 			continue;
 		}
 		if (rw == DDIR_READ)
@@ -130,7 +130,7 @@
 		else if (rw == DDIR_WRITE)
 			writes++;
 		else {
-			fprintf(stderr, "bad ddir: %d\n", rw);
+			fprintf(f_err, "bad ddir: %d\n", rw);
 			continue;
 		}
 
@@ -169,7 +169,7 @@
 		return 0;
 
 	if (td->rate < td->ratemin) {
-		fprintf(stderr, "min rate larger than nominal rate\n");
+		fprintf(f_err, "min rate larger than nominal rate\n");
 		return -1;
 	}
 
diff --git a/stat.c b/stat.c
index a540304..cfd27ce 100644
--- a/stat.c
+++ b/stat.c
@@ -233,7 +233,7 @@
 		p = dirname(foo);
 		sprintf(tmp, "%s/queue", p);
 		if (stat(tmp, &st)) {
-			fprintf(stderr, "unknown sysfs layout\n");
+			fprintf(f_err, "unknown sysfs layout\n");
 			return;
 		}
 		sprintf(foo, "%s", p);
@@ -287,12 +287,12 @@
 
 static void show_group_stats(struct group_run_stats *rs, int id)
 {
-	printf("\nRun status group %d (all jobs):\n", id);
+	fprintf(f_out, "\nRun status group %d (all jobs):\n", id);
 
 	if (rs->max_run[DDIR_READ])
-		printf("   READ: io=%lluMiB, aggrb=%llu, minb=%llu, maxb=%llu, mint=%llumsec, maxt=%llumsec\n", rs->io_kb[0] >> 10, rs->agg[0], rs->min_bw[0], rs->max_bw[0], rs->min_run[0], rs->max_run[0]);
+		fprintf(f_out, "   READ: io=%lluMiB, aggrb=%llu, minb=%llu, maxb=%llu, mint=%llumsec, maxt=%llumsec\n", rs->io_kb[0] >> 10, rs->agg[0], rs->min_bw[0], rs->max_bw[0], rs->min_run[0], rs->max_run[0]);
 	if (rs->max_run[DDIR_WRITE])
-		printf("  WRITE: io=%lluMiB, aggrb=%llu, minb=%llu, maxb=%llu, mint=%llumsec, maxt=%llumsec\n", rs->io_kb[1] >> 10, rs->agg[1], rs->min_bw[1], rs->max_bw[1], rs->min_run[1], rs->max_run[1]);
+		fprintf(f_out, "  WRITE: io=%lluMiB, aggrb=%llu, minb=%llu, maxb=%llu, mint=%llumsec, maxt=%llumsec\n", rs->io_kb[1] >> 10, rs->agg[1], rs->min_bw[1], rs->max_bw[1], rs->min_run[1], rs->max_run[1]);
 }
 
 static void show_disk_util(void)
@@ -302,7 +302,7 @@
 	struct disk_util *du;
 	double util;
 
-	printf("\nDisk stats (read/write):\n");
+	fprintf(f_out, "\nDisk stats (read/write):\n");
 
 	list_for_each(entry, &disk_list) {
 		du = list_entry(entry, struct disk_util, list);
@@ -312,7 +312,7 @@
 		if (util > 100.0)
 			util = 100.0;
 
-		printf("  %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, in_queue=%u, util=%3.2f%%\n", du->name, dus->ios[0], dus->ios[1], dus->merges[0], dus->merges[1], dus->ticks[0], dus->ticks[1], dus->time_in_queue, util);
+		fprintf(f_out, "  %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, in_queue=%u, util=%3.2f%%\n", du->name, dus->ios[0], dus->ios[1], dus->merges[0], dus->merges[1], dus->ticks[0], dus->ticks[1], dus->time_in_queue, util);
 	}
 }
 
@@ -328,19 +328,19 @@
 		return;
 
 	bw = td->io_bytes[ddir] / td->runtime[ddir];
-	printf("  %s: io=%6lluMiB, bw=%6lluKiB/s, runt=%6lumsec\n", ddir_str[ddir], td->io_bytes[ddir] >> 20, bw, td->runtime[ddir]);
+	fprintf(f_out, "  %s: io=%6lluMiB, bw=%6lluKiB/s, runt=%6lumsec\n", ddir_str[ddir], td->io_bytes[ddir] >> 20, bw, td->runtime[ddir]);
 
 	if (calc_lat(&td->slat_stat[ddir], &min, &max, &mean, &dev))
-		printf("    slat (msec): min=%5lu, max=%5lu, avg=%5.02f, dev=%5.02f\n", min, max, mean, dev);
+		fprintf(f_out, "    slat (msec): min=%5lu, max=%5lu, avg=%5.02f, dev=%5.02f\n", min, max, mean, dev);
 
 	if (calc_lat(&td->clat_stat[ddir], &min, &max, &mean, &dev))
-		printf("    clat (msec): min=%5lu, max=%5lu, avg=%5.02f, dev=%5.02f\n", min, max, mean, dev);
+		fprintf(f_out, "    clat (msec): min=%5lu, max=%5lu, avg=%5.02f, dev=%5.02f\n", min, max, mean, dev);
 
 	if (calc_lat(&td->bw_stat[ddir], &min, &max, &mean, &dev)) {
 		double p_of_agg;
 
 		p_of_agg = mean * 100 / (double) rs->agg[ddir];
-		printf("    bw (KiB/s) : min=%5lu, max=%5lu, per=%3.2f%%, avg=%5.02f, dev=%5.02f\n", min, max, p_of_agg, mean, dev);
+		fprintf(f_out, "    bw (KiB/s) : min=%5lu, max=%5lu, per=%3.2f%%, avg=%5.02f, dev=%5.02f\n", min, max, p_of_agg, mean, dev);
 	}
 }
 
@@ -352,7 +352,7 @@
 	if (!(td->io_bytes[0] + td->io_bytes[1]) && !td->error)
 		return;
 
-	printf("%s: (groupid=%d): err=%2d:\n",td->name, td->groupid, td->error);
+	fprintf(f_out, "%s: (groupid=%d): err=%2d:\n",td->name, td->groupid, td->error);
 
 	show_ddir_status(td, rs, td->ddir);
 	if (td->io_bytes[td->ddir ^ 1])
@@ -368,7 +368,7 @@
 		sys_cpu = 0;
 	}
 
-	printf("  cpu          : usr=%3.2f%%, sys=%3.2f%%, ctx=%lu\n", usr_cpu, sys_cpu, td->ctx);
+	fprintf(f_out, "  cpu          : usr=%3.2f%%, sys=%3.2f%%, ctx=%lu\n", usr_cpu, sys_cpu, td->ctx);
 }
 
 void show_run_stats(void)
@@ -393,7 +393,7 @@
 		td = &threads[i];
 
 		if (td->error) {
-			printf("%s: %s\n", td->name, td->verror);
+			fprintf(f_out, "%s: %s\n", td->name, td->verror);
 			continue;
 		}