The empirical identification of non-linearities in investment relies on how investment is assumed to be separated into various regimes. Using German establishment-level panel data, we estimate a two-regime model of replacement and expansion investment which allows us to observe regime separation, an aspect of the data that is typically absent from previous empirical studies. Our results indicate that firms tend to spread the expansion of capital stock over a period of years rather than concentrating investment in a single year. Moreover, there is evidence that investment is more sensitive to fundamentals in the high regime, where establishments both replace and expand capital stock, than in the low regime, where they only invest in replacement. Finally, correcting for endogenous sample selection indicates that this source of bias does not affect the coefficient estimates significantly.