diff --git a/src/structs/dmat.rs b/src/structs/dmat.rs
index 9a2bc2ca..d2c86acd 100644
--- a/src/structs/dmat.rs
+++ b/src/structs/dmat.rs
@@ -2,7 +2,7 @@
 
 #![allow(missing_docs)] // we hide doc to not have to document the $trhs double dispatch trait.
 
-use std::{cmp, mem};
+use std::cmp;
 use std::iter::repeat;
 use std::ops::{Add, Sub, Mul, Div, Index, IndexMut};
 use std::fmt::{Debug, Display, Formatter, Result};
@@ -480,16 +480,15 @@ impl<N: Clone + Copy> Transpose for DMat<N> {
     #[inline]
     fn transpose_mut(&mut self) {
         if self.nrows == self.ncols {
-            for i in 1..self.nrows {
-                for j in 0..self.ncols - 1 {
+            let n = self.nrows;
+            for i in 0..n - 1 {
+                for j in i + 1..n {
                     let off_i_j = self.offset(i, j);
                     let off_j_i = self.offset(j, i);
 
                     self.mij[..].swap(off_i_j, off_j_i);
                 }
             }
-
-            mem::swap(&mut self.nrows, &mut self.ncols);
         }
         else {
             // FIXME: implement a better algorithm which does that in-place.
diff --git a/tests/mat.rs b/tests/mat.rs
index f4c70f83..234beba1 100644
--- a/tests/mat.rs
+++ b/tests/mat.rs
@@ -3,7 +3,7 @@ extern crate rand;
 
 use rand::random;
 use na::{Vec1, Vec3, Mat1, Mat2, Mat3, Mat4, Mat5, Mat6, Rot2, Rot3, Persp3, PerspMat3, Ortho3,
-         OrthoMat3, DMat, DVec, Row, Col, BaseFloat, Diag};
+         OrthoMat3, DMat, DVec, Row, Col, BaseFloat, Diag, Transpose, RowSlice};
 
 macro_rules! test_inv_mat_impl(
   ($t: ty) => (
@@ -688,4 +688,19 @@ fn test_cholesky_mat5() {
 #[test]
 fn test_cholesky_mat6() {
     test_cholesky_impl!(Mat6<f64>);
-}
\ No newline at end of file
+}
+
+#[test]
+fn test_transpose_square_mat() {
+    let col_major_mat = &[0, 1, 2, 3,
+                          0, 1, 2, 3,
+                          0, 1, 2, 3,
+                          0, 1, 2, 3];
+    let num_rows = 4;
+    let num_cols = 4;
+    let mut mat = DMat::from_col_vec(num_rows, num_cols, col_major_mat);
+    mat.transpose_mut();
+    for i in 0..num_rows {
+        assert_eq!(&[0, 1, 2, 3], mat.row_slice(i, 0, num_cols).as_slice());
+    }
+}