The effect of spatial variations in mean radiant temperature and occupancy on HVAC energy and occupant thermal comfort is studied in an operi-plan office space with multiple AHUs. A predictive control method is proposed that arrives at optimum temperature set-point vector by solving an optimization problem, which minimizes HVAC energy consumption subject to acceptable thermal comfort and adequate outside air intake. The optimum temperature set-point vector is used in a PID controller that modulates the AHU fan speed. The proposed control is evaluated on occupancy traces observed in an open-plan space and compared with static set-points based PID control strategies. Normally PID control operates with a dead band around the static set-point and the AHU fan speed is adjusted only when return air temperature is outside the dead-band. The proposed control has a dynamic set-point and hence the AHU speeds are adjusted at much finer time steps than the normal control. The occupancy traces are classified into densely and sparsely occupied days. Across all days, the proposed Control achieves an average additional savings of 15% over a PID control that assumes uniform spatial occupancy distribution in AHU control and 12% over a PID based strategy that uses actual spatial occupancy information. (C) 2017 Elsevier B.V. All rights reserved.