Computer-based assessments open up new possibilities to measure constructs in authentic settings. They are especially promising to measure 21st century skills, as for instance information and communication technologies (ICT) skills. Items tapping such constructs may be diverse regarding design principles and content and thus form a heterogeneous item set. Existing validation approaches, as the construct representation approach by Embretson (1983), however, require homogenous item sets in the sense that a particular task characteristic can be applied to all items. To apply this validation rational also for heterogeneous item sets, two experimental approaches are proposed based on the idea to create variants of items by systematically manipulating task characteristics. The change-approach investigates whether the manipulation affects construct-related demands and the eliminate-approach whether the test score represents the targeted skill dimension. Both approaches were applied within an empirical study (N = 983) using heterogeneous items from an ICT skills test. The results show how changes of ICT-specific task characteristics influenced item difficulty without changing the represented construct. Additionally, eliminating the intended skill dimension led to easier items and changed the construct partly. Overall, the suggested experimental approaches provide a useful validation tool for 21st century skills assessed by heterogeneous items.