The objective of this paper is to investigate if and how capital adjustment departs from the smooth pattern implied by standard model based on convex adjustment costs. Using Norwegian micro data, we start by documenting various aspects of the distribution of investment rates. We then present two pieces of econometric evidence on these issues. First, we estimate a discrete hazard model to determine the probability of having an episode of high investment, conditional on the length of the interval from the last high investment episode and we discuss what the empirical results suggest about the shape of the adjustment cost function. Second, we move beyond this discretization of the investment problem and estimate a switching regression model that allows for the response of the investment rate to fundamentals to differ across regimes. In both cases we investigate the aggregate implications of our results.