Three studies compared different computer generated voices for helicopter voice warnings. Study 1 compared three LPC-encoded voices (human female, human male, and phoneme synthesized) while pilots flew simulated nap-of-the earth missions. Flight performance and pilots' ratings via semantic differential scales were measured. Study 2 produced pilot preferences for direct synthesized speech compared to LPC-encoded human female speech and compared to LPC-encoded synthesized speech. Study 3 used phonetically balanced (PB) words heard in simulated helicopter noise to compare intelligibility of direct synthesized and LPC-encoded phoneme synthesized speech. There were no flight performance differences due to voice type (Study 1), but pilots preferred direct synthesized speech over both LPC-encoded human female speech and LPC-encoded synthesized speech (Study 2). PB word intelligibility was better for direct synthesized speech than for LPC-encoded synthesized speech; there were no differences associated with sex of listener or with being a pilot or not. A runs by speech type interaction suggests differential learnability of the two speech types. (Study 3). Implications of these findings for voice selection will be discussed.