Purpose: The lexical selection hypothesis posits that first words added to a toddler's spoken vocabulary will be predominantly those beginning with early developing consonant phonemes. Using this framework, we evaluated the relationship between word form and lexical selection among late talkers and two typical comparison groups. Method: An online database of MacArthur-Bates Communicative Developmental Inventories was used to extract the American English Words and Sentences Form (MB-CDI:WS). Inventories were divided into three groups: (a) a late talkers group (LTs; n = 202), (b) a typically developing age-matched group (TDA; n = 1,238), and (c) a younger, typically developing language-matched group (TDL; n = 196) matched on expressive language to the LTs. The first phoneme in each word produced by every toddler on the MB-CDI:WS was coded as early, middle, or late developing. The proportion of spoken words starting with phonemes in each developmental category was calculated. Mixed-effects models were used to evaluate group differences. Results: All three groups' spoken vocabularies consisted mostly of words beginning with early developing phonemes. LTs and TDLs used more words beginning with early developing consonants than TDAs. TDAs had a higher proportion of words starting with middle- and late- developing phonemes than LTs and TDL groups. The LTs group produced a significantly smaller proportion of words beginning with middle-developing phonemes compared to the TDL group. Conclusions: Initial phonemes produced in the lexicons of LTs are, in general, similar to both language-matched and age-matched typical toddlers and reflect lexical selection. Clinical implications of these findings will be discussed.