This paper deals with on-line identification and constrained long-range predictive control of multivariable systems. It extends a recently proposed augmented upper diagonal factorization identification (AUDI) algorithm to identify input-output models of multivariable systems with distinct time delays. The multi-input, multi-output (MIMO AUDI) algorithm can simultaneously identify the process model order and process parameters. The MIMO AUDI algorithm is implemented by decomposing a MIMO system into as many multi-input, single-output (MISO) subsystems as the number of outputs and then identifying each MISO subsystem separately. The performance of the new MIMO AUDI algorithm is demonstrated by application to input-output data from a real process. The extension of this algorithm by incorporating a variable forgetting factor with a lower bound in its value is implemented on real plant data to demonstrate 'alertness' of the estimator. This paper evaluates the performance of the MIMO adaptive generalized predictive control algorithm with and without constraints by experimental application on a computer-interfaced, pilot-scale process. The MIMO adaptive GPC is shown to have good regulatory plus servo-tracking properties.