We explore the link between short-run cycles and long-run growth by examining the cyclicality of R&D. Existing theories propose that R&D is concentrated when output is low, but aggregate data repeatedly show that R&D appears procyclical. We estimate the relationship between R&D and output using an annual panel of twenty U.S. manufacturing industries from 1958 to 1998. The results indicate that R&D is in fact procyclical, but, interestingly, estimates using demand-shift instruments suggest that it responds asymmetrically to demand shocks. We propose that liquidity constraint is an important cause for the observed procyclicality of R&D.