Dental students commit many errors when diagnosing radiographs. To improve performance, students were asked to compare radiographs (with and without disease or with the same disease); relevant structures were highlighted in the radiographs. In a crossover design, students were randomly assigned to two groups differing in training order: Students in the peripheral-central-group (N = 39) were first trained to detect anomalies in the periphery before receiving training on anomalies in the center; the trainings in the central-peripheral-group (N = 39) were reversed. We measured detection rates and gaze behavior before and after each training. The detection rates after the first training revealed differences in line with our expectations; moreover, when accounting for varying difficulty of the tests sets there were within-groups improvements in the peripheral-central group. Unexpectedly, the gaze behavior was unaffected by the intervention. We discuss shorter learning times and sequence effects as potential causes for our findings.