We estimate a model of homogeneous capital investment with two installation possibilities – replacement and expansion using observations at the establishment level. We find that regime switches identified by ad hoc measures of lumpy investment do not adequately distinguish expansionary from replacement activities. In fact, during periods of expansion, firms spend just as much on replacement capital. Also, using the common 20% rule would not assign a spike to almost 65% of all observations that include expansionary investment in this dataset. Finally, replacement although less responsive to fundamentals than expansions cannot be regarded as an autonomous part of investment.