We studied whether magnetoencephalography (MEG) could detect deceptive responses on a single-subject, trial-by-trial basis. To elicit spontaneous, ecologically valid deception, we developed a paradigm in which subjects in a simulated customs setting were presented with a series of pictures of items which might be in their baggage, and for each item, they decided whether to "declare" (tell the truth) or "smuggle" (lie). Telling the truth involved a small but certain monetary penalty, whereas lying involved both greater monetary risk and greater potential reward. Most subjects showed decreased signal power in the 8-12 Hz (alpha) range during deceptive responses as compared to truthful responses. In a cross-validation analysis, we were able to use alpha power to classify truthful and deceptive responses on a trial-by-trial basis, with significantly greater predictive accuracy than that achieved using simultaneously recorded skin conductance signals. Average predictive accuracy for spontaneous deception was greater than 78%, and for some subjects, predictive accuracy exceeded 90%. Our results raise the possibility that alpha power modulation during deception may reflect risk management and/or cognitive control.