Friction force is a crucial factor causing power loss and fatigue spalling of rolling element bearings. A combined experimental and analytical method is proposed to quantitatively determine the elastohydrodynamic lubrication (EHL) friction force distribution between rollers and outer raceway in a cylindrical roller bearing (CRB). An experimental system with the instrumented bearing and housing was developed for measuring radial load distribution and friction torque of bearings. A simplified model of friction force expressed by dimensionless speed, load, and material parameters was given. An inequality constrained optimization problem was established and solved by using an experimental data-driven learning algorithm for determining the uncertain parameters in the model. The effect of speed, load, and lubricant property on friction force and friction coefficient was discussed.