Masked priming was used to study the locus of the semantic interference effect. This effect was studied by varying the nature of the task (naming and gender decision) and the type of prime-target relation (semantic and phonological). The time of presentation of the prime was also varied. Results indicated that semantic interference appears in naming (Experiment 1) and gender decision (Experiment 2) when the time of presentation of the prime was 100 ms. This replicates results by others and extends them to gender decision. In contrast, phonological facilitation was not present in gender decision (Experiment 3). This pattern suggests that semantic interference is the result of processes occurring at the lemma level and that gender decision is not influenced by phonological activation.