In three experiments, we evaluated potential sources of familiarity in the production of feature and conjunction errors with both word (Experiments 1 and 3) and nonword (Experiment 2) stimuli and related this work to various dual-process models of memory. The contributions of letter, syllable, lexical morpheme, and conceptual information were considered. Lexical morpheme information was consistently more potent than syllable information in leading to feature and conjunction errors across Experiments 1 and 2, and a word length explanation did not account for this consistent finding. In addition, there was no impact of conceptual information on these errors (Experiments 1-3). The results support a familiarity-based interpretation of feature and conjunction errors and a lexical morpheme basis for the familiarity in compound words. In order to be more comprehensive, memory models may need to account for a lexical morpheme source of familiarity.