A case study effort to model distributed computer-aided engineering environments is presented. Typical environments will be hosted on graphics workstations communicating over a local-area-network and supported by one or more file-servers. The paper describes our initial attempts at quantifying performance of alternate architectures being proposed for these engineering environments. Specifically, we focus on predicting workstation performance as a function of local memory. The modeling approach includes Generalized Stochastic Petri Nets (GSPN), Multiclass (BCMP) Queueing Networks, and an experimental method of measuring program characteristics for input parameter estimation.