The paper analyses on an experimental basis the phenomenon of non-optimal under-diversification in portfolio choice decisions and investigates the reasons behind it. The most important obstacles for optimal diversification are studied the correlation neglect hypothesis and the overconfidence which both lead to suboptimal diversification decisions. The investment alternatives are constructed in a way that the Markowitz' efficiency frontier is reduced to a single point in the return-risk diagram so that unambiguous interpretation of the results is possible: the subjects neglect the correlation between the assets, use naïve diversification strategies and take irrelevant information as a foundation for their investment decisions, the first effect being stronger than the second.