Hodrick-Prescott filter has been the favourite empirical technique among researchers studying "cycles". Software facilities and the optimality criterion, from which the filter can be derived, can explain its wide use. However, different shortcomings and drawbacks have been pointed out in the literature, as alteration of variability and persistence and detecting spurious cycles and correlations. This paper discusses these critics from an empirical point of view trying to clarify what the filter can and cannot do. In particular, a less mechanical use for descriptive analysis is proposed: testing how the estimated cyclical component behaves and using autocorrelation adjusted standard errors to evaluate cross correlations to differentiate the "genuine" from "spurious" case. Simulation results to test these bivariate correlations when there is a "genuine" relationship are presented. Some examples of descriptive analysis for macro aggregates (real activity, trade flows and money) of Argentina are reported to show that not always the filter is appropriate and simple tools could be used to appreciate how the filtered series result and to evaluate cross correlations.