Living the economic dream of globalization in the form of a location and time-independent world-wide employment market, today crowd sourcing companies offer affordable digital solutions to business problems. At the same time, highly accessible economic opportunities are offered to workers, who often live in low or middle income countries. Thus, crowd sourcing can be understood as a flexible social solution that indiscriminately reaches out to poor, yet diligent workers: a win-win situation for employers and crowd workers. On the other hand, its virtual nature opens doors to unethical exploitation by fraudulent workers, compromising in turn the overall quality of the gained results and increasing the costs of continuous result quality assurance, e.g. by gold questions or majority votes. The central question discussed in this paper is how to distinguish between basically honest workers, who might just be lacking educational skills, and plainly unethical workers. We show how current quality control measures misjudge and subsequently discriminate against honest workers with lower skill levels. In contrast, our techniques use statistical models that computes the level of a worker’s skill and a task’s difficulty to clearly distinguish each worker’s success zone and detect irrational response patterns, which usually imply fraud. Our evaluation shows that about 50% of misjudged workers can be successfully detected as honest, can be retained, and subsequently redirected to easier tasks.