diff --git a/nalgebra-sparse/src/ops/serial/coo.rs b/nalgebra-sparse/src/ops/serial/coo.rs
deleted file mode 100644
index 322c6914..00000000
--- a/nalgebra-sparse/src/ops/serial/coo.rs
+++ /dev/null
@@ -1,72 +0,0 @@
-//! Matrix operations involving sparse matrices.
-
-use crate::coo::CooMatrix;
-use nalgebra::base::storage::{Storage, StorageMut};
-use nalgebra::{ClosedAdd, ClosedMul, Dim, Scalar, Vector};
-use num_traits::{One, Zero};
-
-/// Sparse matrix-vector multiplication `y = beta * y + alpha * A * x`.
-///
-/// Computes a matrix-vector product with the COO matrix "A" and the vector `x`, storing the
-/// result in `y`.
-///
-/// If `beta == 0`, the elements in `y` are never read.
-///
-/// TODO: Rethink this function
-///
-/// Panics
-/// ------
-///
-/// Panics if `y`, `a` and `x` do not have compatible dimensions.
-pub fn spmv_coo<T, Y, X, YDim, XDim>(
-    beta: T,
-    y: &mut Vector<T, YDim, Y>,
-    alpha: T,
-    a: &CooMatrix<T>,
-    x: &Vector<T, XDim, X>,
-) where
-    T: Scalar + ClosedAdd + ClosedMul + Zero + One,
-    YDim: Dim,
-    XDim: Dim,
-    Y: StorageMut<T, YDim>,
-    X: Storage<T, XDim>,
-{
-    assert_eq!(
-        y.len(),
-        a.nrows(),
-        "y and a must be dimensionally compatible"
-    );
-    assert_eq!(
-        a.ncols(),
-        x.len(),
-        "a and x must be dimensionally compatible"
-    );
-
-    if beta == T::zero() {
-        // If `y` is constructed through `new_uninitialized()`, we must make sure to not read
-        // any of the elements in order to avoid UB, so we special case beta == 0
-        // in order to ensure that we only write, not read, the elements in y.
-        for y_i in y.iter_mut() {
-            *y_i = T::zero();
-        }
-    } else if beta != T::one() {
-        // Since the COO triplets have no particular structure, we cannot combine initialization
-        // of y with the triplet loop below, and instead have to do it in a pre-pass.
-        for y_i in y.iter_mut() {
-            *y_i *= beta.inlined_clone();
-        }
-    }
-
-    for (i, j, v) in a.triplet_iter() {
-        // TODO: We could skip bounds checks with unsafe here, since COO ensures that all indices
-        // are in bounds and we assert on dimensions up-front.
-        // The compiler will not be able to elide the checks, since we're doing
-        // random/unpredictable access to elements in `x` and `y`.
-        let (alpha, v, x_j) = (
-            alpha.inlined_clone(),
-            v.inlined_clone(),
-            x[j].inlined_clone(),
-        );
-        y[i] += alpha * v * x_j;
-    }
-}
diff --git a/nalgebra-sparse/src/ops/serial/mod.rs b/nalgebra-sparse/src/ops/serial/mod.rs
index cd0dc09c..01c31d93 100644
--- a/nalgebra-sparse/src/ops/serial/mod.rs
+++ b/nalgebra-sparse/src/ops/serial/mod.rs
@@ -30,11 +30,9 @@ macro_rules! assert_compatible_spmm_dims {
     }
 }
 
-mod coo;
 mod csr;
 mod pattern;
 
-pub use coo::*;
 pub use csr::*;
 pub use pattern::*;
 
diff --git a/nalgebra-sparse/tests/unit_tests/ops.rs b/nalgebra-sparse/tests/unit_tests/ops.rs
index 6c4f2006..a7f82b9a 100644
--- a/nalgebra-sparse/tests/unit_tests/ops.rs
+++ b/nalgebra-sparse/tests/unit_tests/ops.rs
@@ -1,11 +1,10 @@
-use nalgebra_sparse::coo::CooMatrix;
-use nalgebra_sparse::ops::serial::{spmv_coo, spmm_csr_dense, spadd_build_pattern, spadd_csr};
+use nalgebra_sparse::ops::serial::{spmm_csr_dense, spadd_build_pattern, spadd_csr};
 use nalgebra_sparse::ops::{Transpose};
 use nalgebra_sparse::csr::CsrMatrix;
 use nalgebra_sparse::proptest::{csr, sparsity_pattern};
 use nalgebra_sparse::pattern::SparsityPattern;
 
-use nalgebra::{DVector, DMatrix, Scalar, DMatrixSliceMut, DMatrixSlice};
+use nalgebra::{DMatrix, Scalar, DMatrixSliceMut, DMatrixSlice};
 use nalgebra::proptest::matrix;
 
 use proptest::prelude::*;
@@ -24,31 +23,6 @@ fn dense_csr_pattern(pattern: &SparsityPattern) -> DMatrix<i32> {
     DMatrix::from(&boolean_csr)
 }
 
-#[test]
-fn spmv_coo_agrees_with_dense_gemv() {
-    let x = DVector::from_column_slice(&[2, 3, 4, 5]);
-
-    let i = vec![0, 0, 1, 1, 2, 2];
-    let j = vec![0, 3, 0, 1, 1, 3];
-    let v = vec![3, 2, 1, 2, 3, 1];
-    let a = CooMatrix::try_from_triplets(3, 4, i, j, v).unwrap();
-
-    let betas = [0, 1, 2];
-    let alphas = [0, 1, 2];
-
-    for &beta in &betas {
-        for &alpha in &alphas {
-            let mut y = DVector::from_column_slice(&[2, 5, 3]);
-            let mut y_dense = y.clone();
-            spmv_coo(beta, &mut y, alpha, &a, &x);
-
-            y_dense.gemv(alpha, &DMatrix::from(&a), &x, beta);
-
-            assert_eq!(y, y_dense);
-        }
-    }
-}
-
 #[derive(Debug)]
 struct SpmmCsrDenseArgs<T: Scalar> {
     c: DMatrix<T>,