We considered the problem of estimating the frequency of an unknown periodic function from observations which consist of a particular almost periodic function and additive Gaussian white noise from both theoretical and practical points of view. First, we proved that, in a precise asymptotic framework, our estimator was consistent and asymptotically efficient. Then, when the number of observations was fixed, we proposed a practical implementation of the procedure and compared its performance with those forecast by the theoretical study. Finally, we applied our processing to some laser vibrometry signals and compared it with a technique classically used in this field. Copyright 2006 Blackwell Publishing Ltd.