Climatic characteristics are affected by various systematic and occasional impacts: besides the changes in the observing system (locations of the stations of the meteorological network, instruments, observing procedures), the possible local-scale and global natural and antropogenic impacts on climatic conditions should be taken into account. Apart from the predictability problems, the phenomenological analysis of the climatic variability and the determination of past persistent climatic anomalies are significant problems, among other aspects, as evidence of the possible anomalous behavior of climate or for climate impact studies. In this paper, a special technique for the identification of such "shifts" in the observational series is presented. The existence of these significant shorter or longer term changes in the mean characteristics for the properly selected adjoining periods of time is the necessary condition for the formation of any more or less unidirectional climatic trends. Actually, the window technique is based on a complete set of orthogonal functions. The sensitivity of the proposed model on its main parameters is also investigated. This method is applied for hemispheric and Hungarian data series of the mean annual surface temperature.