In three ERP experiments, morphology-based decomposition of words and pseudowords was explored in Spanish. Subjects were asked to perform a lexical decision task on morphologically simple (e.g. sun) and complex (e.g. allerg+ic, allerg+ist) word strings, while family size for both lexemes/stems (S-FS) and morphemes/suffixes (M-FS) was varied. In Experiment I, earlier results by Schreuder & Baayen (1997) were replicated: Monomorphemic High-FS targets produced faster responses than monomorphemic Low-FS targets. On the whole, mean voltage amplitudes for Low FS were higher than for High FS; the actual process takes place, however, at a late stage. In Experiment II, where lexical roots were used as primes, late Family Size effects emerged. Lexical decisions were based principally on lexicality. In Experiment III, words and pseudowords were matched for Stem Family Size and Morpheme Family Size was manipulated: morphemes were used as primes. Word and pseudoword targets which contained a high family size (HFS) morpheme require 21ms less than word and pseudoword targets which did not, showing that HFS morphemes facilitate word recognition. Our data with monomorphemic words seems to support Schreuder & Baayen's (1995) model. Results with polymorphemic words cannot be accommodated within a framework that ignores the relationship between whole word and morpheme.