diff --git a/include/mlir/IR/Statements.h b/include/mlir/IR/Statements.h
index 46abe6f..4fe49bd 100644
--- a/include/mlir/IR/Statements.h
+++ b/include/mlir/IR/Statements.h
@@ -251,7 +251,10 @@
   void setUpperBoundMap(AffineMap *map);
 
   /// Set loop step.
-  void setStep(int64_t step) { this->step = step; }
+  void setStep(int64_t step) {
+    assert(step > 0 && "step has to be a positive integer constant");
+    this->step = step;
+  }
 
   /// Returns true if the lower bound is constant.
   bool hasConstantLowerBound() const;
@@ -342,7 +345,8 @@
   AffineMap *lbMap;
   // Affine map for the upper bound.
   AffineMap *ubMap;
-  // Constant step.
+  // Positive constant step. Since affineint is int64_t, we restrict step to the
+  // set of positive integers that int64_t can represent.
   int64_t step;
   // Operands for the lower and upper bounds, with the former followed by the
   // latter. Dimensional operands are followed by symbolic operands for each
diff --git a/lib/Analysis/LoopAnalysis.cpp b/lib/Analysis/LoopAnalysis.cpp
index fa283ae..f4e607e 100644
--- a/lib/Analysis/LoopAnalysis.cpp
+++ b/lib/Analysis/LoopAnalysis.cpp
@@ -66,13 +66,12 @@
 
     auto *cExpr = dyn_cast<AffineConstantExpr>(loopSpanExpr);
     if (!cExpr)
-      return AffineBinaryOpExpr::getCeilDiv(loopSpanExpr, std::abs(step),
-                                            context);
+      return AffineBinaryOpExpr::getCeilDiv(loopSpanExpr, step, context);
     loopSpan = cExpr->getValue();
   }
 
   // 0 iteration loops.
-  if ((loopSpan < 0 && step >= 1) || (loopSpan > 0 && step <= -1))
+  if (loopSpan < 0)
     return 0;
 
   return AffineConstantExpr::get(
diff --git a/lib/IR/Statement.cpp b/lib/IR/Statement.cpp
index c4eb5b8..7834d32 100644
--- a/lib/IR/Statement.cpp
+++ b/lib/IR/Statement.cpp
@@ -268,6 +268,7 @@
          "lower bound operand count does not match the affine map");
   assert(ubOperands.size() == ubMap->getNumInputs() &&
          "upper bound operand count does not match the affine map");
+  assert(step > 0 && "step has to be a positive integer constant");
 
   unsigned numOperands = lbOperands.size() + ubOperands.size();
   ForStmt *stmt =
diff --git a/lib/Parser/Parser.cpp b/lib/Parser/Parser.cpp
index f9c801c..54ddb7a 100644
--- a/lib/Parser/Parser.cpp
+++ b/lib/Parser/Parser.cpp
@@ -2277,6 +2277,13 @@
   if (consumeIf(Token::kw_step) && parseIntConstant(step))
     return ParseFailure;
 
+  // The loop step is a positive integer constant. Since affineint is of int64_t
+  // type, we restrict step to be in the set of positive integers that int64_t
+  // can represent.
+  if (step < 1) {
+    return emitError("step has to be a positive integer");
+  }
+
   // Create for statement.
   ForStmt *forStmt =
       builder.createFor(getEncodedSourceLocation(loc), lbOperands, lbMap,
diff --git a/test/IR/invalid.mlir b/test/IR/invalid.mlir
index 05557af..3e962f5 100644
--- a/test/IR/invalid.mlir
+++ b/test/IR/invalid.mlir
@@ -191,6 +191,12 @@
 
 // -----
 
+mlfunc @for_negative_stride() {
+  for %i = 1 to 10 step -1
+}        // expected-error {{step has to be a positive integer}}
+
+// -----
+
 mlfunc @non_statement() {
   asd   // expected-error {{custom op 'asd' is unknown}}
 }
diff --git a/test/IR/parser.mlir b/test/IR/parser.mlir
index c8c2237..49a7cad 100644
--- a/test/IR/parser.mlir
+++ b/test/IR/parser.mlir
@@ -214,8 +214,8 @@
   %s = "foo"(%N) : (affineint) -> affineint
   // CHECK: for %i0 = %0 to %arg0
   for %i = %s to %N {
-    // CHECK: for %i1 = #map1(%i0) to 0 step -1
-    for %j = %i to 0 step -1 {
+    // CHECK: for %i1 = #map1(%i0) to 0
+    for %j = %i to 0 step 1 {
        // CHECK: %1 = affine_apply #map{{.*}}(%i0, %i1)[%0]
        %w = affine_apply(d0, d1)[s0] -> (d0+d1, s0+1) (%i, %j) [%s]
        // CHECK: for %i2 = #map{{.*}}(%1#0, %i0)[%arg0] to #map{{.*}}(%1#1, %i1)[%0] {