Two lexical decision experiments examining semantic priming from attended and unattended parafoveal words are reported. On every trial, a peripheral cue appearing to the left or right of fixation was followed by two parafoveal primes, one on each visual field, and a subsequent word/nonword target at fixation. In Experiment 1, duration of the cue was manipulated. Parafoveal semantic priming was found but was significant only for the attended primes, showing a similar magnitude of priming for the different cue durations levels. However, there was a great variation in priming among subjects, with some showing positive priming and others showing negative priming. In Experiment 2, different instructions to process the primes were manipulated. Results showed that (a) instructions to attend the primes were associated with positive priming; (b) instructions to ignore the primes were associated with negative priming, even when the ignored primes were cued. Implications of these findings regarding influence of selective attention on semantic processing of parafoveal words are discussed.