This paper studies the dynamic relationship between input and output of innovation in Dutch manufacturing using an unbalanced panel of enterprise data from five waves of the Community Innovation Survey during 1994-2004. We estimate by maximum likelihood a dynamic panel data bivariate tobit with double-index sample selection accounting for individual effects. We find persistence of innovation input and innovation output, a lag effect of the former on the latter and a feedback effect of the latter on the former. The lag effect remains significant in the high-tech sector even after four years. Firm and industry effects are also important.