This paper considers estimation of parameters for a model currently being used to study the Illinois Interlibrary Loan Network (ILLINET) and presents a method for determining how much data should be collected and how it should be aggregated. Four classes of model parameters discussed: demands, probabilities of success, processing times, and delivery times. This approach is being used as a means of deciding when to terminate data collection as well as determining the number of request classes to use in the policy analyses being performed with a library network model. This paper shows how model parameter uncertainty affects prediction uncertainty as a function of sample size and number of request classes. As sample size increases, uncertainty decreases. However, as the number of request classes increases, the sample size in each class decreases and thus, uncertainty increases. An appendix presents a new version of the model used in previous studies which could be applicable to many different types of networks. (DAG)