Multi-variable model predictive control (MMPC) was used to control the dimethyl ether (DME) purification process of methanol in the production of DME from synthesis gas. The use of MMPC aims to capture the phenomenon of the interaction between the variables in the process in order to improve the control performance. As the process comprises four input variables and four output variables, MMPC (4x4) is used in this study. The inter-variable interaction is shown in a 4x4 matrix, where each matrix element is a first-order plus dead-time (FOPDT) model. MMPC (4x4) was tested by changing the set point (SP) and disturbance rejection. The control performance indicators used are integral absolute error (IAE) and integral square error (ISE) and, as a comparison, the control performance of the single-input single-output (SISO) model predictive control (MPC). The results show that MMPC (4x4) is better than MPC in both IAE and ISE. In terms of SP change, MMPC (4x4) is able to significantly improve the control performance of MPC, by 78% (IAE) and 90% (ISE). Whereas in the disturbance rejection testing, the improvements in control performance were 58 % (IAE) and 81% (ISE).