The incorporation of experience curves has enhanced the treatment of technological change in models used to evaluate the cost of climate and energy policies. However, the set of activities that experience curves are assumed to capture is much broader than the set that can be characterized by learning-by-doing, the primary connection between experience curves and economic theory. How accurately do experience curves describe observed technological change? This study examines the case of photovoltaics (PV), a potentially important climate stabilization technology with robust technology dynamics. Empirical data are assembled to populate a simple engineering-based model identifying the most important factors affecting the cost of PV over the past three decades. The results indicate that learning from experience only weakly explains change in the most important cost-reducing factors - plant size, module efficiency, and the cost of silicon. They point to other explanatory variables to include in future models. Future work might also evaluate the potential for efficiency gains from policies that rely less on riding down the learning curve and more on creating incentives for firms to make investments in the types of cost-reducing activities quantified in this study.