The benefits of prior information about who would speak, where they would be located, and when they would speak were measured in a multi-talker spatial-listening task. On each trial, a target phrase and several masker phrases were allocated to 13 loudspeakers in a 180° arc, and to 13 overlapping time slots, which started every 800 ms. Speech-reception thresholds (SRTs) were measured as the level of target relative to masker phrases at which listeners reported key words at 71% correct. When phases started in pairs all three cues were beneficial (“who” 3.2 dB, “where” 5.1 dB, and “when” 0.3 dB). Over a range of onset asynchronies, SRTs corresponded consistently to a signal-to-noise ratio (SNR) of −2 dB at the start of the target phrase. When phrases started one at a time, SRTs fell to a SNR of −8 dB and were improved significantly, but only marginally, by constraining “who” (1.9 dB), and not by constraining “where” (1.0 dB) or “when” (0.01 dB). Thus, prior information about “who,” “where,” and “when” was beneficial, but only when talkers started speaking in pairs. Low SRTs may arise when talkers start speaking one at a time because of automatic orienting to phrase onsets and/or the use of loudness differences to distinguish target from masker phrases.