The differences between historic rainfall data and synthetic data obtained by a stochastic rainfall model are investigated using nonlinear analysis tools devised for description and characterization of chaotic behavior. To achieve this goal, a 6-year point rainfall record with a time resolution of one quarter of hour is studied. A stochastic model capable of preserving important properties of the rainfall process, such as intermittency, seasonality and scaling behavior, is fitted to this data set and a synthetic time series of equal length is generated. For both data sets the correlation dimension is calculated for various embedding dimensions by the time delay embedding method. However, the applicability of this method in estimating dimensions proves limited due to the domination of voids (dry periods) in a rainfall record at a fine time resolution. Thus, in addition to time delay embedding, a Cantorian dust analogue method is developed and used to estimate dimensions. Results of both methods show that there is no substantial difference in behavior between the synthetic and the historic records. Moreover, no evidence of low-dimensional determinism is detected in the sets examined. ¿ American Geophysical Union 1996