In this paper, we consider the indirect adaptive control of processes subject to large, infrequent changes in the process parameters. We attempt to improve the estimation and control by adopting a Gaussian-sum model that fits the characteristics of the parameter changes - better than the usual Gaussian stochastic description leading to the least squares. The optimal estimator that gives the minimum mean-squared error estimate of the parameters has the attractive properties of selectively tuning the estimator gain matrix for different types of parameter changes and being robust to measurement noises. The optimal estimator, however, is computationally infeasible; hence a suboptimal approach that retains the attractive properties of the optimal estimator is proposed. We evaluate the performance of the proposed multivariable adaptive controller by applying it to chemical processes modeled by fundamental and input/output models, with single- and multiple-parameter variations.