This article sums up results from two experimental studies conducted in the context of a research project on the design and assessment of a graphical authentication system (HSA® for Human Semantic Authentication). The three experiments address different issues related to the usability of the system (performance), and its cognitive demands (memorisation and categorisation). The results put into evidence that, despite the fact that this type of system presents some advantages in terms of use (providing some requirements are observed), it raises nonetheless different problems which put uncertainty on its potential acceptabilty and use in real world situations. The results illustrate the traditional tension between security and usability.