Different thermal and non-thermal recovery methods for heavy oil and bitumen extraction are aimed at alternating the in-situ fluid properties of the hydrocarbon phase. Amongst these, viscosity and density reduction are the most favorable ones, which can be achieved through alternation of temperature, pressure, and composition (mixing with various solvents) of the oil. Therefore, proper phase behavior modelling of hydrocarbon phase over a representative range of principle parameters (temperature, pressure, composition), plays a central role in the accuracy of numerical modelling of heavy oil and bitumen recovery schemes. As such, the phase behavior modelling requires the characterization of oil and the tuning of an equation of state which can account for solubility, density, and viscosity variations; all of these three require abundant amount of experimental data. The lack of experimental data for different bitumen/solvent mixtures is occasionally the motive for using default or hypothesized values in empirical correlations which generate these properties in the simulation models. This can lead to unreliable simulation results due to a large deviation from realistic values if not entirely wrong phase properties predictions. Thus, as a part of our project on comprehensive phase behavior study of heavy oils and bitumen, we have focused on gathering, compilation, and finally analysis of almost all available literature data for such hydrocarbons. In the current study, the results of analysis on about 3200 measurements of density (for bitumen and mixtures of bitumen and liquid solvents) in the form of correlation coefficients and summary tables for individual and grouped datasets are presented. A robust optimization algorithm which minimizes the least square error between experimental data and correlation predictions is the main tool in achieving this goal.