diff --git a/benches/bench1.rs b/benches/bench1.rs
index 0486bd092..a6fa86deb 100644
--- a/benches/bench1.rs
+++ b/benches/bench1.rs
@@ -14,6 +14,7 @@ use std::mem::MaybeUninit;
 use ndarray::ShapeBuilder;
 use ndarray::{arr0, arr1, arr2, azip, s};
 use ndarray::{Array, Array1, Array2, Axis, Ix, Zip};
+use ndarray::{Ix1, Ix2, Ix3, Ix5, IxDyn};
 
 use test::black_box;
 
@@ -941,3 +942,59 @@ fn sum_axis1(bench: &mut test::Bencher) {
     let a = range_mat(MEAN_SUM_N, MEAN_SUM_N);
     bench.iter(|| a.sum_axis(Axis(1)));
 }
+
+#[bench]
+fn into_dimensionality_ix1_ok(bench: &mut test::Bencher) {
+    let a = Array::<f32, _>::zeros(Ix1(10));
+    let a = a.view();
+    bench.iter(|| a.into_dimensionality::<Ix1>());
+}
+
+#[bench]
+fn into_dimensionality_ix3_ok(bench: &mut test::Bencher) {
+    let a = Array::<f32, _>::zeros(Ix3(10, 10, 10));
+    let a = a.view();
+    bench.iter(|| a.into_dimensionality::<Ix3>());
+}
+
+#[bench]
+fn into_dimensionality_ix3_err(bench: &mut test::Bencher) {
+    let a = Array::<f32, _>::zeros(Ix3(10, 10, 10));
+    let a = a.view();
+    bench.iter(|| a.into_dimensionality::<Ix2>());
+}
+
+#[bench]
+fn into_dimensionality_dyn_to_ix3(bench: &mut test::Bencher) {
+    let a = Array::<f32, _>::zeros(IxDyn(&[10, 10, 10]));
+    let a = a.view();
+    bench.iter(|| a.clone().into_dimensionality::<Ix3>());
+}
+
+#[bench]
+fn into_dimensionality_dyn_to_dyn(bench: &mut test::Bencher) {
+    let a = Array::<f32, _>::zeros(IxDyn(&[10, 10, 10]));
+    let a = a.view();
+    bench.iter(|| a.clone().into_dimensionality::<IxDyn>());
+}
+
+#[bench]
+fn into_dyn_ix3(bench: &mut test::Bencher) {
+    let a = Array::<f32, _>::zeros(Ix3(10, 10, 10));
+    let a = a.view();
+    bench.iter(|| a.into_dyn());
+}
+
+#[bench]
+fn into_dyn_ix5(bench: &mut test::Bencher) {
+    let a = Array::<f32, _>::zeros(Ix5(2, 2, 2, 2, 2));
+    let a = a.view();
+    bench.iter(|| a.into_dyn());
+}
+
+#[bench]
+fn into_dyn_dyn(bench: &mut test::Bencher) {
+    let a = Array::<f32, _>::zeros(IxDyn(&[10, 10, 10]));
+    let a = a.view();
+    bench.iter(|| a.clone().into_dyn());
+}
diff --git a/src/dimension/dimension_trait.rs b/src/dimension/dimension_trait.rs
index aa1f7f95a..18a10c3f0 100644
--- a/src/dimension/dimension_trait.rs
+++ b/src/dimension/dimension_trait.rs
@@ -929,6 +929,11 @@ impl Dimension for IxDyn {
     fn from_dimension<D2: Dimension>(d: &D2) -> Option<Self> {
         Some(IxDyn(d.slice()))
     }
+
+    fn into_dyn(self) -> IxDyn {
+        self
+    }
+
     private_impl! {}
 }
 
diff --git a/src/impl_methods.rs b/src/impl_methods.rs
index f8c0ee919..341a90286 100644
--- a/src/impl_methods.rs
+++ b/src/impl_methods.rs
@@ -6,6 +6,7 @@
 // option. This file may not be copied, modified, or distributed
 // except according to those terms.
 
+use std::mem::{size_of, ManuallyDrop};
 use alloc::slice;
 use alloc::vec;
 use alloc::vec::Vec;
@@ -1583,8 +1584,11 @@ where
         }
     }
 
-    /// Convert an array or array view to another with the same type, but
-    /// different dimensionality type. Errors if the dimensions don't agree.
+    /// Convert an array or array view to another with the same type, but different dimensionality
+    /// type. Errors if the dimensions don't agree (the number of axes must match).
+    ///
+    /// Note that conversion to a dynamic dimensional array will never fail (and is equivalent to
+    /// the `into_dyn` method).
     ///
     /// ```
     /// use ndarray::{ArrayD, Ix2, IxDyn};
@@ -1600,8 +1604,11 @@ where
     where
         D2: Dimension,
     {
-        if let Some(dim) = D2::from_dimension(&self.dim) {
-            if let Some(strides) = D2::from_dimension(&self.strides) {
+        if D::NDIM == D2::NDIM {
+            // safe because D == D2
+            unsafe {
+                let dim = unlimited_transmute::<D, D2>(self.dim);
+                let strides = unlimited_transmute::<D, D2>(self.strides);
                 return Ok(ArrayBase {
                     data: self.data,
                     ptr: self.ptr,
@@ -1609,6 +1616,17 @@ where
                     strides,
                 });
             }
+        } else if D::NDIM == None || D2::NDIM == None { // one is dynamic dim
+            if let Some(dim) = D2::from_dimension(&self.dim) {
+                if let Some(strides) = D2::from_dimension(&self.strides) {
+                    return Ok(ArrayBase {
+                        data: self.data,
+                        ptr: self.ptr,
+                        dim,
+                        strides,
+                    });
+                }
+            }
         }
         Err(ShapeError::from_kind(ErrorKind::IncompatibleShape))
     }
@@ -2375,3 +2393,18 @@ where
         });
     }
 }
+
+
+/// Transmute from A to B.
+///
+/// Like transmute, but does not have the compile-time size check which blocks
+/// using regular transmute in some cases.
+///
+/// **Panics** if the size of A and B are different.
+#[inline]
+unsafe fn unlimited_transmute<A, B>(data: A) -> B {
+    // safe when sizes are equal and caller guarantees that representations are equal
+    assert_eq!(size_of::<A>(), size_of::<B>());
+    let old_data = ManuallyDrop::new(data);
+    (&*old_data as *const A as *const B).read()
+}