Using autoregressive modeling of discrete signals, we investigate the influence of mass and size on the memory function of a tracer particle immersed in a Lennard-Jones liquid. We find that the memory function of the tracer particle scales with the inverse reduced mass of the simulated system. Increasing the particle's mass leads rapidly to a slow exponential decay of the velocity autocorrelation function, whereas the memory function changes just its amplitude. This effect is the more pronounced the smaller and the heavier the tracer particle is. (C) 2003 American Institute of Physics.