Rational choice theory predicts for higher punishment less crime. However, many field studies could not support this conclusion. A game theoretic approach can explain these puzzling findings because it takes not only criminals' but also control agents' rationality into account. Mixed Nash equilibria predict for higher punishment less control and no effect on crime rates. A new experimental design is introduced to test game theoretic hypotheses. 196 subjects have been partitioned into `inspectees' who can steal money from each other and `inspectors' who can invest in control activities to catch inspectees. Static and dynamic analyses show that strategic interaction plays an important role for crime and punishment. However, effects are not as strong as predicted. Higher punishment indeed causes less control, but crime is deterred as well. Furthermore, dynamical analyses with the learning model fictitious play reveal that humans learn only slowly in inspection situations.