The two-part estimation technique has been advocated for estimating models using individual level health care utilization data characterised by a large proportion of non-consumption and small proportions of heavy users. This paper compares the two-part model to several other estimators, including the Poisson, negative binomial and 'zero altered' negative binomial models on the basis of within-sample forecasting accuracy and non-nested model selection tests. The empirical model estimates the differential effect of the removal of copayments for prescription medicines on the prescription drug utilization by older adults with differing levels of health status. The two-part estimator of this model is found to dominate the competitors. Results from this model indicate that utilization increases appear to be higher among individuals with lower levels of health status.