Theory of Mind (ToM) represents a key aspect of human intelligence, but it is still unclear whether Artificial Intelligence (AI) can learn this ability. Previous works attempted to test the ToM ability on AI models by using different implementations like text or images but none of them did follow a Visual Question Answering (VQA) approach. This work presents the new data set CLEVR-ToM, which for the first time represents false-belief tests as VQA tasks. By using a VQA approach, it addresses two important human senses with natural language (text) and visual (image) information. Especially for the Sally-Anne test, which tests a location false-belief, this VQA version appears very beneficial as it shows many similarities to the original form of the test. For the testing, this work extends the CNN+LSTM+RN model to a new model CNN+2LSTM+RN to better fit the new CLEVR-ToM data set. The CNN+2LSTM+RN model delivered outstanding results on the CLEVR-ToM data set with an accuracy of almost 98%, achieving higher results than the original model. This work proves for the first time that it is possible to implement the false-belief test in a VQA fashion and that the models can handle the tasks very well. This lays the foundation for further tests of other, even more challenging ToM types, that can be built on this basis.