Evidence for cognitive benefits of simulated versus physical experiments is unclear. Seventh grade participants (n = 147) reported their understanding of two simple pendulum problems (1) before conducting an experiment, (2) immediately following experimentation, and (3) after a 12-week delay. "Problem type" was manipulated within subjects--participants' understanding of one problem was typically accurate and for the other was typically inaccurate. Experiments were computer-simulated or hands-on and were observed in a slow motion or real time replay. There was no difference between simulated vs. hands-on or slow motion vs. real time replay for conceptual understanding outcomes. Instead, the problem type, the time of posttest, and the participant's experimentation strategy were significant predictors. Specifically, poor experiments were especially bad at informing understanding of the previously misconceived problem. Furthermore, even for the problem that was previously conceived correctly, accurate understanding declined over twelve weeks for those participants who conducted inadequate experiments. Post hoc tests revealed participants were less likely to control variables during simulations. (Contains 3 tables and 4 figures.)