Hydraulic tests are widely used for qualitative and quantitative characterization of aquifers. They allow understanding the conceptual model and estimating the parameters that govern the water flow by recording and interpreting the aquifer response to pumping/injection. This response needs to be represented in terms of drawdown, defined as the change in head caused by the pumping test. As such, they are calculated as the difference between the head that would have been observed without pumping (i.e., “natural” head) and the head that has been actually observed. The problem lies on how to evaluate the natural head. That is, heads monitored during a hydraulic test are usually affected by regional trends, recharge events and more generally, by perturbations other than those of the test itself. These perturbations need to be filtered from the time series to isolate the effect of pumping. Filtering is usually carried out by means of statistical methods (e.g. Fourier-based filtering, average removing method and adaptive detrending algorithm). Our work is motivated by long tests where the aquifer recent past may affect head evolution during the test. We propose a detrending methodology to be applied for hydrological series in complex systems, which consist on the development and application of two flow numerical models, jointly with the stochastic inversion, as a single tool for measured data filtering. We proceed iteratively between the two models, so that the most recent calibrated field is used for modeling natural head evolution and for calculating drawdowns. The method is illustrated by application to a synthetic and a real case.