diff --git a/Cargo.toml b/Cargo.toml
index 736b9738..fe87f723 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -51,9 +51,18 @@ pest_derive    = { version = "2.0", optional = true }
 [dev-dependencies]
 serde_json = "1.0"
 rand_xorshift = "0.1"
+criterion = "0.2"
 
 [workspace]
 members = [ "nalgebra-lapack", "nalgebra-glm" ]
 
+[[bench]]
+name = "nalgebra_bench"
+harness = false
+path = "benches/lib.rs"
+
 [patch.crates-io]
 alga = { path = "../alga/alga" }
+
+[profile.bench]
+lto = true
diff --git a/benches/common/macros.rs b/benches/common/macros.rs
index 758336a8..43bcf59d 100644
--- a/benches/common/macros.rs
+++ b/benches/common/macros.rs
@@ -2,56 +2,52 @@
 
 macro_rules! bench_binop(
     ($name: ident, $t1: ty, $t2: ty, $binop: ident) => {
-        #[bench]
-        fn $name(bh: &mut Bencher) {
+        fn $name(bh: &mut criterion::Criterion) {
             use rand::SeedableRng;
             let mut rng = IsaacRng::seed_from_u64(0);
             let a = rng.gen::<$t1>();
             let b = rng.gen::<$t2>();
 
-            bh.iter(|| {
+            bh.bench_function(stringify!($name), move |bh| bh.iter(|| {
                 a.$binop(b)
-            })
+            }));
         }
     }
 );
 
 macro_rules! bench_binop_ref(
     ($name: ident, $t1: ty, $t2: ty, $binop: ident) => {
-        #[bench]
-        fn $name(bh: &mut Bencher) {
+        fn $name(bh: &mut criterion::Criterion) {
             use rand::SeedableRng;
             let mut rng = IsaacRng::seed_from_u64(0);
             let a = rng.gen::<$t1>();
             let b = rng.gen::<$t2>();
 
-            bh.iter(|| {
+            bh.bench_function(stringify!($name), move |bh| bh.iter(|| {
                 a.$binop(&b)
-            })
+            }));
         }
     }
 );
 
 macro_rules! bench_binop_fn(
     ($name: ident, $t1: ty, $t2: ty, $binop: path) => {
-        #[bench]
-        fn $name(bh: &mut Bencher) {
+        fn $name(bh: &mut criterion::Criterion) {
             use rand::SeedableRng;
             let mut rng = IsaacRng::seed_from_u64(0);
             let a = rng.gen::<$t1>();
             let b = rng.gen::<$t2>();
 
-            bh.iter(|| {
+            bh.bench_function(stringify!($name), move |bh| bh.iter(|| {
                 $binop(&a, &b)
-            })
+            }));
         }
     }
 );
 
 macro_rules! bench_unop_na(
     ($name: ident, $t: ty, $unop: ident) => {
-        #[bench]
-        fn $name(bh: &mut Bencher) {
+        fn $name(bh: &mut criterion::Criterion) {
             const LEN: usize = 1 << 13;
 
             use rand::SeedableRng;
@@ -60,21 +56,20 @@ macro_rules! bench_unop_na(
             let elems: Vec<$t> =  (0usize .. LEN).map(|_| rng.gen::<$t>()).collect();
             let mut i = 0;
 
-            bh.iter(|| {
+            bh.bench_function(stringify!($name), move |bh| bh.iter(|| {
                 i = (i + 1) & (LEN - 1);
 
                 unsafe {
                     test::black_box(na::$unop(elems.get_unchecked(i)))
                 }
-            })
+            }));
         }
     }
 );
 
 macro_rules! bench_unop(
     ($name: ident, $t: ty, $unop: ident) => {
-        #[bench]
-        fn $name(bh: &mut Bencher) {
+        fn $name(bh: &mut criterion::Criterion) {
             const LEN: usize = 1 << 13;
 
             use rand::SeedableRng;
@@ -83,21 +78,20 @@ macro_rules! bench_unop(
             let mut elems: Vec<$t> =  (0usize .. LEN).map(|_| rng.gen::<$t>()).collect();
             let mut i = 0;
 
-            bh.iter(|| {
+            bh.bench_function(stringify!($name), move |bh| bh.iter(|| {
                 i = (i + 1) & (LEN - 1);
 
                 unsafe {
                     test::black_box(elems.get_unchecked_mut(i).$unop())
                 }
-            })
+            }));
         }
     }
 );
 
 macro_rules! bench_construction(
     ($name: ident, $constructor: path, $( $args: ident: $types: ty),*) => {
-        #[bench]
-        fn $name(bh: &mut Bencher) {
+        fn $name(bh: &mut criterion::Criterion) {
             const LEN: usize = 1 << 13;
 
             use rand::SeedableRng;
@@ -106,14 +100,14 @@ macro_rules! bench_construction(
             $(let $args: Vec<$types> = (0usize .. LEN).map(|_| rng.gen::<$types>()).collect();)*
             let mut i = 0;
 
-            bh.iter(|| {
+            bh.bench_function(stringify!($name), move |bh| bh.iter(|| {
                 i = (i + 1) & (LEN - 1);
 
                 unsafe {
                     let res = $constructor($(*$args.get_unchecked(i),)*);
                     test::black_box(res)
                 }
-            })
+            }));
         }
     }
 );
diff --git a/benches/core/matrix.rs b/benches/core/matrix.rs
index c323cd6e..896f2c97 100644
--- a/benches/core/matrix.rs
+++ b/benches/core/matrix.rs
@@ -196,3 +196,5 @@ fn mat100_from_fn(bench: &mut Bencher) {
 fn mat500_from_fn(bench: &mut Bencher) {
     bench.iter(|| DMatrix::from_fn(500, 500, |a, b| a + b))
 }
+
+criterion_group!(matrix, mat2_transpose);
\ No newline at end of file
diff --git a/benches/core/mod.rs b/benches/core/mod.rs
index 9699a728..f3e633a8 100644
--- a/benches/core/mod.rs
+++ b/benches/core/mod.rs
@@ -1,2 +1,2 @@
-mod matrix;
-mod vector;
+pub mod matrix;
+pub mod vector;
diff --git a/benches/core/vector.rs b/benches/core/vector.rs
index 837eb7ca..814f50e7 100644
--- a/benches/core/vector.rs
+++ b/benches/core/vector.rs
@@ -1,7 +1,6 @@
 use na::{DVector, Vector2, Vector3, Vector4, VectorN};
 use rand::{IsaacRng, Rng};
 use std::ops::{Add, Div, Mul, Sub};
-use test::{self, Bencher};
 use typenum::U10000;
 
 #[path = "../common/macros.rs"]
@@ -48,19 +47,17 @@ bench_unop!(vec4_normalize, Vector4<f32>, normalize);
 bench_binop_ref!(vec10000_dot_f64, VectorN<f64, U10000>, VectorN<f64, U10000>, dot);
 bench_binop_ref!(vec10000_dot_f32, VectorN<f32, U10000>, VectorN<f32, U10000>, dot);
 
-#[bench]
-fn vec10000_axpy_f64(bh: &mut Bencher) {
+fn vec10000_axpy_f64(bh: &mut criterion::Criterion) {
     use rand::SeedableRng;
     let mut rng = IsaacRng::seed_from_u64(0);
     let mut a = DVector::new_random(10000);
     let b = DVector::new_random(10000);
     let n = rng.gen::<f64>();
 
-    bh.iter(|| a.axpy(n, &b, 1.0))
+    bh.bench_function("", move |bh| bh.iter(|| a.axpy(n, &b, 1.0)));
 }
 
-#[bench]
-fn vec10000_axpy_beta_f64(bh: &mut Bencher) {
+fn vec10000_axpy_beta_f64(bh: &mut criterion::Criterion) {
     use rand::SeedableRng;
     let mut rng = IsaacRng::seed_from_u64(0);
     let mut a = DVector::new_random(10000);
@@ -68,27 +65,25 @@ fn vec10000_axpy_beta_f64(bh: &mut Bencher) {
     let n = rng.gen::<f64>();
     let beta = rng.gen::<f64>();
 
-    bh.iter(|| a.axpy(n, &b, beta))
+    bh.bench_function("", move |bh| bh.iter(|| a.axpy(n, &b, beta)));
 }
 
-#[bench]
-fn vec10000_axpy_f64_slice(bh: &mut Bencher) {
+fn vec10000_axpy_f64_slice(bh: &mut criterion::Criterion) {
     use rand::SeedableRng;
     let mut rng = IsaacRng::seed_from_u64(0);
     let mut a = DVector::new_random(10000);
     let b = DVector::new_random(10000);
     let n = rng.gen::<f64>();
 
-    bh.iter(|| {
+    bh.bench_function("", move |bh| bh.iter(|| {
         let mut a = a.fixed_rows_mut::<U10000>(0);
         let b = b.fixed_rows::<U10000>(0);
 
         a.axpy(n, &b, 1.0)
-    })
+    }));
 }
 
-#[bench]
-fn vec10000_axpy_f64_static(bh: &mut Bencher) {
+fn vec10000_axpy_f64_static(bh: &mut criterion::Criterion) {
     use rand::SeedableRng;
     let mut rng = IsaacRng::seed_from_u64(0);
     let mut a = VectorN::<f64, U10000>::new_random();
@@ -96,22 +91,20 @@ fn vec10000_axpy_f64_static(bh: &mut Bencher) {
     let n = rng.gen::<f64>();
 
     // NOTE: for some reasons, it is much faster if the arument are boxed (Box::new(VectorN...)).
-    bh.iter(|| a.axpy(n, &b, 1.0))
+    bh.bench_function("", move |bh| bh.iter(|| a.axpy(n, &b, 1.0)));
 }
 
-#[bench]
-fn vec10000_axpy_f32(bh: &mut Bencher) {
+fn vec10000_axpy_f32(bh: &mut criterion::Criterion) {
     use rand::SeedableRng;
     let mut rng = IsaacRng::seed_from_u64(0);
     let mut a = DVector::new_random(10000);
     let b = DVector::new_random(10000);
     let n = rng.gen::<f32>();
 
-    bh.iter(|| a.axpy(n, &b, 1.0))
+    bh.bench_function("", move |bh| bh.iter(|| a.axpy(n, &b, 1.0)));
 }
 
-#[bench]
-fn vec10000_axpy_beta_f32(bh: &mut Bencher) {
+fn vec10000_axpy_beta_f32(bh: &mut criterion::Criterion) {
     use rand::SeedableRng;
     let mut rng = IsaacRng::seed_from_u64(0);
     let mut a = DVector::new_random(10000);
@@ -119,5 +112,7 @@ fn vec10000_axpy_beta_f32(bh: &mut Bencher) {
     let n = rng.gen::<f32>();
     let beta = rng.gen::<f32>();
 
-    bh.iter(|| a.axpy(n, &b, beta))
+    bh.bench_function("", move |bh| bh.iter(|| a.axpy(n, &b, beta)));
 }
+
+criterion_group!(vector, vec10000_axpy_f64_static);
diff --git a/benches/lib.rs b/benches/lib.rs
index f788c998..7c8cdc70 100644
--- a/benches/lib.rs
+++ b/benches/lib.rs
@@ -6,15 +6,20 @@ extern crate rand;
 extern crate test;
 extern crate typenum;
 
+#[macro_use]
+extern crate criterion;
+
 use na::DMatrix;
 use rand::{IsaacRng, Rng};
 
-mod core;
-mod geometry;
-mod linalg;
+pub mod core;
+pub mod geometry;
+pub mod linalg;
 
 fn reproductible_dmatrix(nrows: usize, ncols: usize) -> DMatrix<f64> {
     use rand::SeedableRng;
     let mut rng = IsaacRng::seed_from_u64(0);
     DMatrix::<f64>::from_fn(nrows, ncols, |_, _| rng.gen())
 }
+
+criterion_main!(core::matrix::matrix, core::vector::vector);
\ No newline at end of file