We perform an incentivized experiment designed to assess the accuracy of beliefs about characteristics and decisions. Subjects are asked to declare some specific choices and characteristics with different levels of observability from an external point of view, and typically formed through real world experiences. From the less observable mobile phone purchasing decisions, hypothetical restaurant choices, political views, happiness to the fully observable height and weight; after they are asked to report beliefs on statistics over the same items concerning other individuals living in the same environment. We test two main hypotheses: (i) whether for items not perfectly observable, individuals suffer of some type of bias in these beliefs; (ii) whether this bias would disappear for weight and height, when the information is perfectly available. We find a powerful and ubiquitous bias in perceptions that is "self-centered" in the sense that those at extremes tend to perceive themselves as closer to the middle of the distribution than is the case. Albeit weaker, this bias does not disappear when the information is readily available as in height and weight. We present evidence from our experiment that limited attention and self-serving deception can provide explanations for this bias and present important economic applications.