diff --git a/lib/Conversion/ArithToNeura/ArithToNeuraPass.cpp b/lib/Conversion/ArithToNeura/ArithToNeuraPass.cpp
index dc6f4532..cec3f673 100644
--- a/lib/Conversion/ArithToNeura/ArithToNeuraPass.cpp
+++ b/lib/Conversion/ArithToNeura/ArithToNeuraPass.cpp
@@ -57,6 +57,28 @@ struct ArithAddIToNeuraAdd : public OpRewritePattern<mlir::arith::AddIOp> {
   }
 };
 
+struct ArithCmpFToNeuraFCmp : public OpRewritePattern<mlir::arith::CmpFOp> {
+  using OpRewritePattern::OpRewritePattern;
+
+  LogicalResult matchAndRewrite(arith::CmpFOp op,
+                                PatternRewriter &rewriter) const override {
+  
+    Value lhs = op.getLhs();
+    Value rhs = op.getRhs();
+
+    mlir::arith::CmpFPredicate predicate_enum = op.getPredicate();
+
+    StringRef predicate_str = arith::stringifyCmpFPredicate(predicate_enum);
+    
+    StringAttr predicate_attr = rewriter.getStringAttr(predicate_str);
+// Converts arith CmpFOp to Neura FCmpOp.
+    rewriter.replaceOpWithNewOp<neura::FCmpOp>(
+        op, op.getResult().getType(), lhs, rhs, predicate_attr);
+
+    return success();
+  }
+};
+
 struct ArithFAddToNeuraFAdd : public OpRewritePattern<mlir::arith::AddFOp> {
   using OpRewritePattern::OpRewritePattern;
 
@@ -338,7 +360,7 @@ struct LowerArithToNeuraPass
           mlir::neura::arith2neura::populateWithGenerated(patterns);
           patterns.add<
               ArithFAddToNeuraFAdd, ArithConstantToNeuraConstant,
-              ArithAddIToNeuraAdd, ArithCmpiToNeuraICmp, ArithSelectToNeuraSel,
+              ArithAddIToNeuraAdd, ArithCmpiToNeuraICmp, ArithCmpFToNeuraFCmp, ArithSelectToNeuraSel,
               ArithExtUIToNeuraCast, ArithIndexCastToNeuraCast,
               ArithFDivToNeuraFDiv, ArithExtfToNeuraCast, ArithMulFToNeuraFMul,
               ArithSubIToNeuraSub, ArithSubFToNeuraFSub, ArithMulIToNeuraMul,
diff --git a/test/arith2neura/cmpf.mlir b/test/arith2neura/cmpf.mlir
new file mode 100644
index 00000000..27b25eca
--- /dev/null
+++ b/test/arith2neura/cmpf.mlir
@@ -0,0 +1,12 @@
+module {
+  func.func @test_cmpf(%arg0: f32, %arg1: f32) -> i1 {
+    %0 = arith.cmpf ogt, %arg0, %arg1 : f32
+    return %0 : i1
+  }
+}
+
+// RUN: mlir-neura-opt --assign-accelerator --lower-arith-to-neura %s | FileCheck %s --check-prefix=OPT
+
+// CHECK-LABEL: func.func @test_cmpf(
+// OPT: %{{.*}} = "neura.fcmp"
+// OPT: cmpType = "ogt"