The paper introduces a methodology for measuring accurately the time it takes respondents to answer questions in computer-assisted telephone surveys. The methodology, which is completely invisible to respondents, comprises a computer “clock,” capable of timing responses with millisecond accuracy, and a “voice-key” that converts sounds emitted by respondents into signals capable of triggering the computer clock. Response times to questions on a range of attitude questions, including stable and unstable attitudes, were measured. The results revealed orderly norms in the latencies associated with various types of survey questions. The latencies associated with the expression of stable and unstable attitudes are discussed in the context of the notion of “nonattitudes” and shed new light on this controversial thesis. Overall, our results demonstrate that response latencies can be measured precisely and reliably in telephone surveys and that the data from such measurement open new windows on the cognitive dynamics of survey responses.