Computational methods that predict tissue assembly aid in the production of in vitro constructs that mimic native tissue. In particular prostate cancer cells self-assemble on an attachment-limiting substrate into spheroids that resemble micrometastases. These tissue constructs exhibit drug resistance that approaches clinical levels and have application to high-throughput drug testing and design of patient-specific treatments. Two mathematical models of spheroid formation have been developed based on collision theory and Monte Carlo technique. The models accommodate a variety of size populations in the inoculum: single cells, spheroids of different sizes, and combinations of cells and spheroids. Model simulations provide an excellent fit to experimental concentrations of spheroids as measured by the residual error between these two data sets. Collision theory predicts spheroid size distributions over a 5-fold range of cell concentrations in the inoculum. Also it accurately predicts trends in the adhesion properties of DU 145, LNCaP and PC 3 cells, including an up-regulation in the expression patterns of E-cadherin and other adhesion molecules upon spheroid formation. Monte Carlo simulations predict long-range interactions between aggregating cells on the order of several cell diameters. This study provides the first evidence that cancer cells, which have deficient gap junctions, communicate with intercellular bridges that transport membrane vesicles (1 to 3 microns in diameter) between cells. The bridges contain tubulin and can extend at least 100 microns in length. The computational methods presented here have proven exceptional robust in predicting the physical assembly of spheroids and underlying biological phenomena. Since the composition of spheroids is dependent on their size, the models may be able to predict both spheroid size and composition from the properties of the inoculum.