Adult age differences in automatic and controlled semantic priming were investigated by varying the probability of valid primes in a lexical decision task. Tachistoscopic parafoveal stimulus presentation was used to assess age differences in accuracy and response bias as well as latency. Both age groups showed the expected findings of benefits without costs under automatic priming and benefits and costs under controlled priming. Errors for young adults were distributed equally among word and nonword stimuli, whereas older adults displayed a strong tendency to commit errors on nonword trials.