Video games are becoming increasingly popular, especially in the industry sector and as motivating examples in programming education. However, looking at current practices and available tools reveals a dire need for more research in automated game testing as most companies still test manually and students are left off with no tool support. Many challenges arise when developing automated game testing tools, such as difficult-to-reach program statements, heavy randomisation and an explosion of valid program states, exacerbating the test oracle problem. In the past, neuroevolution was successfully used to optimise networks towards mastering various video game genres. However, networks trained to win games are not optimised to reach program statements reliably such that observed program states can be systematically validated. Thus, this thesis aims to combine neuroevolution with search-based software testing (SBST) to automatically generate test suites of neural networks capable of validating the correctness of games reliably, even in the face of strong program randomisation. Initial results in the Scratch programming domain show that the combination of SBST and Neuroevolution is able to reach advanced program statements, and the evolved networks can serve as probabilistic test oracles. However, more work remains to be done as part of this thesis to improve the effectiveness of the proposed approach.