[en] The application of advanced statistical methods (estimation and testprocedures) in the analysis of sequential material balance data may improve implementation of nuclear materials safeguards in bulk handling facilities. In particular a hypothesis-test has to be developed which allows the detection of a diversion constant in time of small amounts of fissile material over extended periods of time with a high detection probability and a controlled low false alarm probability. It is demonstrated that for the case of independent variables and an infinite time horizon the Power-One-Test yields very satisfactory results. But in the presence of systematic errors of the balance data the requirements with respect to detection - and false alarm - probability can not be met simultaneously. For a finite time horizon the analytical calculation of the false alarm probability proves not to be possible, therefore different methods to estimate the false alarm rate have been studied. Due to the accumulated measurement error the limitations for detection of a protracted diversion can in principle not be influenced, neither by means of intermediate tests nor by application of more complex mathematical evaluation methods or a higher effort for recalibration or multiple measurements. But if it can be assured that within a preceding accountancy interval no material losses will have occured one can adjust for the systematic errors. By that way time periods much longer than that ''clean calibration interval'' can be effectively controlled for any significant material loss. (orig.)