The load-stiffness relationship of a roller sandwiched between two flat surfaces is experimentally investigated. This study provides a comprehensive and well-documented dataset on roller stiffness in bearing applications. The measurements, obtained using hammer impact tests on a variety of rollers are both accurate and precise. Comparing the experimental results with the predictions from the commonly used Tripp model, we observe that the stiffness of the roller under relatively light load conditions is significantly lower than what is predicted by the Tripp model. This discrepancy is primarily attributed to surface roughness, a factor not accounted for in the Tripp model. For practical engineering applications, we propose simple formulas that predict the stiffness of real-world roller-rail contacts with an accuracy of ±10%.