AutoDecon is a fully automatic, multiparameter deconvolution procedure that can be used to estimate various hormone secretion kinetics. We propose a strategy based on the application of the corrected Akaike's information criterion to select the optimal deconvolution model from a class of candidates generated by applying different combinations of initializing values. Using simulated cortisol time series, we show that, particularly in cases of diminished hormone half-life, this approach can yield estimates that are closer to the true underlying secretion kinetics compared with the commonly used approach based on a single set of initializing values. However, although we provide proof of principle, more extensive elaboration and validation of this approach are necessary.