The main objective of this dissertation is to investigate, using Monte Carlo simulations, some statistical properties of GARMA (Generalized Autoregressive Moving Average ) models for time series of count data. GARMA models are extensions of the Generalized Linear Models to dependent data, in which autoregressive (AR) and/or moving average (MA) terms are incorporated into the linear predictor. The statistical properties targeted in our investigation were the model stationarity conditions and the identification criteria for selection of model orders, the lag structure (p,q) associated with the AR and MA terms. Our results suggest that AIC, BIC and Hann-Quinn criteria worked relatively well in identifying the model order, and that the conditions for stationarity established empirically in terms of parameter space restrictions were not totally conclusive, requiring further investigation. As a secondary objective we tested the model against real data, by fitting both a GARMA-Poisson and a GARMA-Negative Binomial to the series of number of cases of poliomyelitis on the US and the number of heart-attacks in Rio de Janeiro city. The results we found indicate that these models were able to explain, in a parsimonious way, the variation of both series.