The binding kinetics recorded for many interactions using BIAcore and IAsys optical biosensors do not fit a simple bimolecular interaction model (A + B<-->AB). Three methods of analysis have been used to derive estimates for kinetic constants from such data:linearization, curve fitting using the integrated rate equation, and curve fitting using numerical integration. To test how well these methods could interpret complex binding kinetics, we generated and analyzed simulated data for two systems, one involving a two-state conformational change (A + B<-->AB<-->(AB)*) and a second involving surface heterogeneity (A + B<-->AB and A + B*<-->AB*). The linearization method assumed a simple bimolecular interaction and was inadequate at interpreting these systems as both produced complex kinetics in the association and dissociation phases. The sum of two integrated rate equations correctly modeled surface heterogeneity; but, when applied nonglobally, it fit the data from the conformational change system equally well and thus provided misleading results. Numerical integration allowed a choice of model for analysis and was therefore the only method capable of returning accurate estimates of rate constants for both complex systems. Global analysis, in combination with numerical integration, provided a stringent test of the assumed model. However, this stringency suggests that its application to experimental systems will require high-quality biosensor data.