diff --git a/src/lib.rs b/src/lib.rs
index 79cb52f..05963e3 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -250,7 +250,7 @@ impl Mnemonic {
 	where
 		R: rand_core::RngCore + rand_core::CryptoRng,
 	{
-		if word_count < MIN_NB_WORDS || word_count % 6 != 0 || word_count > MAX_NB_WORDS {
+		if is_invalid_word_count(word_count) {
 			return Err(Error::BadWordCount(word_count));
 		}
 
@@ -377,7 +377,7 @@ impl Mnemonic {
 	/// Parse a mnemonic in normalized UTF8 in the given language.
 	pub fn parse_in_normalized(language: Language, s: &str) -> Result<Mnemonic, Error> {
 		let nb_words = s.split_whitespace().count();
-		if nb_words < MIN_NB_WORDS || nb_words % 6 != 0 || nb_words > MAX_NB_WORDS {
+		if is_invalid_word_count(nb_words) {
 			return Err(Error::BadWordCount(nb_words));
 		}
 
@@ -558,6 +558,10 @@ impl str::FromStr for Mnemonic {
 	}
 }
 
+fn is_invalid_word_count(word_count: usize) -> bool {
+	word_count < MIN_NB_WORDS || word_count % 3 != 0 || word_count > MAX_NB_WORDS
+}
+
 #[cfg(test)]
 mod tests {
 	use super::*;
@@ -605,6 +609,14 @@ mod tests {
 		let _ = Mnemonic::generate_in_with(&mut rand::thread_rng(), Language::English, 24).unwrap();
 	}
 
+	#[cfg(feature = "rand")]
+	#[test]
+	fn test_generate_word_counts() {
+		for word_count in [12, 15, 18, 21, 24].iter() {
+			let _ = Mnemonic::generate(*word_count).unwrap();
+		}
+	}
+
 	#[test]
 	fn test_vectors_english() {
 		// These vectors are tuples of