This paper studies the persistence of innovation in Dutch manufacturing using an unbalanced panel of firm data from four waves of the Community Innovation Survey between 1994 and 2002. We estimate by maximum likelihood a dynamic type 2 tobit model accounting for individual effects and handling the initial conditions problem. We find true persistence in the probability of innovating in the high-tech category of industries and spurious persistence in the low-tech category. Furthermore, past innovation output intensity affects, albeit to a small extent, current innovation output intensity in the high-tech category, while no such evidence is found in the low-tech category.