Oil recovery in layered reservoirs within a complex geological structure is very often lower than expected. This is due both to the uncertainty in assessing hydraulic continuity between wells and to the trapping of oil in some unswept layers or field areas. The symptoms of such a situation, for a well with commingled production from separate layers inequally waterflooded, may be erratic watercut variations. Such variations may be the result of interferences from surrounding injection and production wells that create changes in rate distribution between the various layers. The objective of this paper is to present a new methodology to model the watercut of a producer as the response of a linear system using neighbouring well rates as inputs. The methodology based on a system identification approach implies the following steps: filtering of the raw data, selection of input wells by a technique of Main Component Analysis, identification of model coefficients by fitting the watercut of the modelled well over a restricted period of production, validation of the identified model and finally evaluation of the proper contribution of each input well. The methodology has been applied to actual data from an area of the Arlan field in Russia. Models obtained were able to successfully forecast the watercut variations of most of the production wells over a several months period. It was then possible to use the geological model of the field area for interpreting these models in terms of main interactions between wells and sweep efficiency for each layer. As shown in the paper, the proposed methodology has the potential (I) to define main hydraulic communications between wells, (II) to give a diagnosis on oil and water saturations of the different pay zones for a commingled producing well and (III) more generally to provide a predictive tool for the interpretation and improvement of the sweeping in layered reservoirs. P. 35