The paper discusses the multivariable modelling and control of a glasshouse micro-climate. A linear reduced-order control model is obtained from a nonlinear simulation model using novel data-based model reduction and linearisation techniques. This control model is then used to design two multivariable non-minimal state variable feedback (SVF) control systems. The first utilises an LQoptimal Proportional-Integral-Plus (PIP) design method incorporating multi-objective optimisation of the weighting matrices, achieving partial dynamic decoupling; while the second uses an algebraic approach to combined pole-assignment and full dynamic decoupling. These controllers are evaluated, to ensure robustness, using the nonlinear simulation model, prior to implementation and evaluation on the real glasshouse during the 1993-94 winter growing season. Control results are excellent with very tight control to the desired setpoints in all three climate variables. For example, air temperature is controlled to within 0.5°C of the setpoint for 85% of the validation period, and is shown to be very robust to model uncertainty and extreme weather conditions.