The main objective of the present thesis is to enhance insight into predictive controller design and identification in connection with steering dynamics. In Chapter 2, the dynamics of ship steering are reviewed. Models of different complexity, suitable for control systems design are presented. The influence of wind, waves and currents on the chip motions are also discussed. Chapter 3 deals with the model reduction problem. Some basic concepts are explained, due to their role in the reduction of the dynamic models. Two model reductions techniques, based on singular values, are described. The theoretical properties of these methods are studied and their performance is examined via simulation on a stochastic linear Mariner Class Vessel model. In Chapter 4, the attention is focused on the derivation of an extended GPC. This extended strategy implies a generalization of the model structure and of the loss function, which defines the optimality of the control. Some guidelines on how to choose the design parameters, depending on the type of process to be controlled and on the required control performance, are presented. A predictive track keeping system for a Mariner Class Vessel is formulated based on the minimization of the mean squares prediction errors of the ship's deviation from the desired track. Chapter 5 is concerned with constrained predictive control. The presented algorithm, which is based on Rosen's gradient projection method, minimizes a multi-step quadratic loss function, taking physical constrints systemat- ically into account. The constraints may consist of amplitude constraints (signal level constraints) as well as rate constraints. The influences of the different parameters on the solution are illustrated via simulation experi- ments on a Mariner Class Vessel model. The results show that the proposed strategy leads to a significant better control than the ad-hoc control strategy. Chapter 6 gives a survey on the so-called forgetting factor methods designed for tracking slowly drifting system parameters. The goal of this cpapter is to formulate the identification framework in order to support the under- standing of the connection between identification and control, analysed in Chapter 7. Chapter 7 focuses on how to make the on-line identification for predictive control more robust towards unmodelled dynamics. The theory is verified via simulation studies on a Mariner Class Vessel. The effects and the need of a prefilter in the estimation are analysed and illustrated. Based on the idea that the control criterion must be dual to the estimation criterion, an iterative optimal prefilter is designed. This seems to be an appealing way to tune the model towards the objective for which the model is to be used.