The objective of this study was to assess changes in semantic priming in two experiments which separately manipulated word length and the location of a semantic incongruence within spoken compound words. All words began 1500 ms after the presentation of related, partially unrelated, or totally unrelated picture-primes. Using a picture of a dogbone as an example, either the spoken words DOGBONE, WISHBONE, DOGHOUSE, or MAILBOX could occur as targets (underscores indicate semantically unrelated parts). In practice, only one of the four possible targets occurred with each prime. Subjects made speeded responses by pressing one of two buttons (related/unrelated) while event-related brain potentials (ERPs) were collected from various scalp locations. The behavioral analyses of response time and accuracy were sensitive to changes in word length, but inconclusive for semantic relatedness. However, repeated-measures ANOVAs of peak amplitude, latency, and mean area amplitude of the N400 ERP revealed significant effects of word length and semantic priming. Comparison of related compounds with partially unrelated word-initial violations of context (e.g., WISHBONE) revealed diminished activity following the peak of the N400. Comparison of partially unrelated word-final contextual violations (e.g., DOGHOUSE) revealed an N400 effect delayed by approximately the mean length of the word-initial component. These results demonstrate the dynamic sensitivity of the cortical semantic processor.