Dynamic Traffic Assignment (DTA) models are important decision support tools for transportation planning and real-time traffic management. One of the biggest obstacles of applying DTA in large-scale networks is the calibration of model parameters, which is essential for the realistic replication of the traffic condition. This paper proposes a methodology for the simultaneous demand-supply DTA calibration based on both aggregate measurements and disaggregate route choice observations to improve the calibration accuracy. The calibration problem is formulated as a bi-level constrained optimization problem and an iterative solution algorithm is proposed. A case study in a highly congested urban area of Beijing using DynaMIT-P is conducted and the combined calibration method improves the fits to surveillance data compared to the calibration based on aggregate measurements only.