Little is known about how firms learn to use new technologies. Using novel data on inputs, profits, and information sets, I study how oil companies learned to use hydraulic fracturing technology in North Dakota between 2005-2012. Firms only partially learned to make profitable input choices, capturing just 60% of possible profits in 2012. To understand why, I estimate a model of input use under technology uncertainty. Firms chose fracking inputs with higher expectations but lower uncertainty about profits, consistent with passive learning but not active experimentation. Most firms over-weighed their own information. These results provide evidence of impediments to learning.