The usage of collaborative filtering mechanisms has become widespread on the web. But how reliable are such recommendation systems as a basis for making out choices? This thesis starts by compiling a list of known weaknesses of rating scales and written reviews, based on existing work within the field of social navigation and a few other disciplines. Four Norwegian web sites, all implementing collaborative filtering techniques, are then analyzed in relation to the compilation of known weaknesses. Based on this analysis, I conclude, with some limitations, that the four cases are highly unreliable as recommendation systems. This conclusion can be extended beyond the four cases studied, since it can make users more critical to the usage of recommendations found on the web, and enable designers to construct more reliable recommendation systems.