Guilty Knowledge Test measuring electrodermal reactions was carried out in order to investigate the quality of different questions and the validity of the test in a situation that resembled a true crime. Fifty participants were randomly assigned to commit one of two realistic mock crimes, and were later tested with GKTs concerning both the crime they had enacted and the one they had no knowledge of. Different scoring systems (SCRs and peak amplitudes as well as raw and standardised scores) were employed and compared when analyzing the results. Although there were some false positives, the test was able to differentiate between the groups of guilty and innocent participants. With the best scoring systems, the test was able to classify up to 84% of the innocent and up to 76% of the guilty correctly according to a logistic regression analysis. ROC areas reflecting these same results reached values above .80. Questions on matters that demanded the participants' attention and were easier to remember had better discriminative power. With nearly all scoring methods, there was a significant interaction between the salience of the relevant items and the guilt of the participants. Participants reacted more strongly to salient relevant items when they were guilty, while no different reactions were observed for the non-salient items between guilty and innocent participants. It is suggested that, although the Guilty Knowledge Test appears to be a valid measure of guilty knowledge even in crimes that are close to real crimes, the principles on which guilty knowledge test questions are constructed should be more clearly specified.