Models of word recognition assume that information about the orthographic form of a word (morpheme) must be available before access to that word's (morpheme's) meaning is possible. In prior work we have demonstrated that semantic similarity influences even early morphological priming (Feldman, Kostić, Gvozdenović, O’Connor, & Martín, 2012; Feldman & Martín, 2009). In two experiments conducted in English, we used a forward-masked lexical decision task to assess whether processing differs after exhaustively decomposable (stem+affix; e.g., pastor-PAST) and partially decomposable (stem+nonmorphemic string; e.g., pasta-PAST) primes in semantically dissimilar prime-target pairs. Results using linear mixed effect models on inverse transformed (-1000/RT) latency data with two separate PCs for the contributions of form (negative) and for frequency (negative) and previous RT as a predictor, failed to show different patterns of facilitation after exhaustively and partially decomposable primes; both of which differed from unrelated controls. Spelling did not interact with prime type but poor spellers varied more across the session.