The recent advance in the development of various hybrid vehicle technologies comes along with the need of establishing optimal energy management strategies, in order to minimize both fuel economy and pollutant emissions, while taking into account an increasing number of state and control variables, depending on the adopted hybrid architecture.One of the objectives of this research was to establish benchmarking performance, in terms of fuel economy, for real time on-board management strategies, such as ECMS (Equivalent Consumption Minimization Strategy), whose structure has been implemented in a SIMULINK model for different hybrid vehicle concepts. The results obtained from these simulations have then been compared with those derived from a general purpose, off-line optimization technique, based on deterministic DP (Dynamic Programming), and capable of finding the numerical global optimum and of generating the optimal cycle-based control trajectory over a discretized multidimensional grid of the selected state and control variables. The paper investigates the structure of the DP problem and its interactions with the specific hybrid architecture, especially in terms of the most appropriate selection of state and control variables. The implications of the chosen modeling approach are also critically evaluated, searching for the best compromise between accurate simulation results and reliable comparisons between off-line and on-line optimization results.One of the outcomes is that the system model should be designed in order to be compatible with efficient DP techniques implementation, with the objective of obtaining robust optimal control policies while achieving acceptable computational costs.The concepts that have been analyzed in this work are the following two parallel hybrid architectures: HEV (Hybrid Electric Vehicle), normally applied in current hybrid vehicles production, and HSF-HV (High Speed Flywheel Hybrid Vehicle), an interesting and promising hybrid mechanical solution. An example of the influence of the selected gear has also been investigated by implementing a multi-dimensional DP optimization routine. In order to perform this analysis, a general purpose DP MATLAB function, including specifically designed algorithms to avoid numerical interpolation issues that typically occur in constrained problems, has been modified to run any SIMULINK-based engine-vehicle model.