Signal detection problems are traditionally viewed as statistical hypothesis testing. In absence of the a priori probabilities, such as in radar, the Neyman-Pearson criterion is used where a certain false alarm probability is set, and the probability of detection is maximised. In signal sensing problems of cognitive radio, the main constraint is to avoid the interference with the primary user. Once this constraint is met, a cognitive radio can maximise its own chance of finding an empty spectrum. In this paper we emphasise this view of the signal sensing problem and modify the criterion such that a maximum miss-detection rate is specified. We have reformulated the energy detector showing that the sensing results have more meaningful explanations under the modified criterion. The effects of measurement errors are also considered.