Uncertainties about technologies and investment opportunities are prevalent for investments in entrepreneurial companies by venture capitalists (VCs), and this study finds that the resolution of these uncertainties, through VCs' learning, is important for their investment decisions. The hypothesis that individual investments are evaluated in isolation, as predicted by standard models, is clearly rejected. The empirical analysis is based on a dynamic learning model derived from the Multi-armed Bandit model. The results suggest that VCs learn from past investments (exploitation) but also consider the option value of future learning (exploration) when making investment decisions.