In this work, a new methodology for modeling qualitative temporal processes, is proposed. In this development, two stages are considered. First, models are obtained and adapted fitting their free parameters according to the existing time series. Second, SelfOrganizing Maps (SOM) are used to establish a clustering of the data using the parameters obtained in the first stage. This methodology is successfully applied to a problem of milk yield prediction in goat herds. Key-Words: Time Series, Neural Networks, Mathematical Models.