diff --git a/Cargo.lock b/Cargo.lock
index 3abdc079..e7a5debc 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -316,7 +316,6 @@ dependencies = [
  "itertools 0.9.0 (registry+https://github.com/rust-lang/crates.io-index)",
  "libdeflater 0.2.0 (registry+https://github.com/rust-lang/crates.io-index)",
  "miniz_oxide 0.3.6 (registry+https://github.com/rust-lang/crates.io-index)",
- "num_cpus 1.13.0 (registry+https://github.com/rust-lang/crates.io-index)",
  "rayon 1.3.0 (registry+https://github.com/rust-lang/crates.io-index)",
  "rgb 0.8.16 (registry+https://github.com/rust-lang/crates.io-index)",
  "wild 2.0.2 (registry+https://github.com/rust-lang/crates.io-index)",
diff --git a/Cargo.toml b/Cargo.toml
index c2b4f672..bcbfa0eb 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -32,7 +32,6 @@ bit-vec = "^0.6.0"
 byteorder = "^1.0.0"
 crc = "^1.2.0"
 itertools = "^0.9.0"
-num_cpus = "^1.13.0"
 zopfli = "^0.4.0"
 miniz_oxide = "0.3"
 rgb = "0.8.11"
diff --git a/src/lib.rs b/src/lib.rs
index 3395510b..9b498728 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -13,7 +13,6 @@
 #![warn(clippy::range_plus_one)]
 #![allow(clippy::cognitive_complexity)]
 
-use num_cpus;
 #[cfg(feature = "parallel")]
 extern crate rayon;
 #[cfg(not(feature = "parallel"))]
@@ -214,10 +213,6 @@ pub struct Options {
     ///
     /// Default: `false`
     pub use_heuristics: bool,
-    /// Number of threads to use
-    ///
-    /// Default: number of CPU cores
-    pub threads: usize,
 
     /// Maximum amount of time to spend on optimizations.
     /// Further potential optimizations are skipped if the timeout is exceeded.
@@ -323,7 +318,6 @@ impl Default for Options {
             strip: Headers::None,
             deflate: Deflaters::Zlib,
             use_heuristics: false,
-            threads: num_cpus::get(),
             timeout: None,
         }
     }
@@ -331,12 +325,6 @@ impl Default for Options {
 
 /// Perform optimization on the input file using the options provided
 pub fn optimize(input: &InFile, output: &OutFile, opts: &Options) -> PngResult<()> {
-    // Initialize the thread pool with correct number of threads
-    #[cfg(feature = "parallel")]
-    let _ = rayon::ThreadPoolBuilder::new()
-        .num_threads(opts.threads)
-        .build_global();
-
     // Read in the file and try to decode as PNG.
     if opts.verbosity.is_some() {
         eprintln!("Processing: {}", input);
@@ -431,12 +419,6 @@ pub fn optimize(input: &InFile, output: &OutFile, opts: &Options) -> PngResult<(
 /// Perform optimization on the input file using the options provided, where the file is already
 /// loaded in-memory
 pub fn optimize_from_memory(data: &[u8], opts: &Options) -> PngResult<Vec<u8>> {
-    // Initialize the thread pool with correct number of threads
-    #[cfg(feature = "parallel")]
-    let _ = rayon::ThreadPoolBuilder::new()
-        .num_threads(opts.threads)
-        .build_global();
-
     // Read in the file and try to decode as PNG.
     if opts.verbosity.is_some() {
         eprintln!("Processing from memory");
diff --git a/src/main.rs b/src/main.rs
index bafdc2bd..8f03e79c 100644
--- a/src/main.rs
+++ b/src/main.rs
@@ -490,7 +490,12 @@ fn parse_opts_into_struct(
     }
 
     if let Some(x) = matches.value_of("threads") {
-        opts.threads = x.parse::<usize>().unwrap();
+        let threads = x.parse::<usize>().unwrap();
+
+        rayon::ThreadPoolBuilder::new()
+            .num_threads(threads)
+            .build_global()
+            .map_err(|err| err.to_string())?;
     }
 
     Ok((out_file, out_dir, opts))