For light curves sampled on an uneven grid of observation times, the shape of the power density spectrum (PDS) includes severe distortion effects due to the window function, and simulations of light curves are indispensable to recover the true PDS. We present an improved method for comparing light curves generated from a PDS model to the measured data and apply it to a 50‐day longRXTEobservations of NGC 4945, a Seyfert 2 galaxy with well‐determined mass from megamaser observations. The improvements over previously reported investigations include the adjustment of the PDS model normalization for each simulated light curve in order to directly investigate how well the chosen PDS shape describes the source data. We furthermore implement a robust goodness‐of‐fit measure that does not depend on the form of the variable used to describe the power in the periodogram. We conclude that a knee‐type function (smoothly broken power law) describes the data better than a simple power law; the best‐fit break frequency is ∼ 10−6Hz. © 2004 American Institute of Physics