Drug discovery relies on our ability to mine the chemical and biological data in an integrated manner. From the initial choice of chemical descriptors to the similarity coefficient itself, each user- and software-made decision will modify the perception of chemical similarity and influence the results of chemical space exploration. Such decisions will alter even the meaning of “similar”, leading to counter-intuitive neighbors. This effort is further complicated by the existence of chemical and target affinity cliffs, which is also noted in the area of marketed drugs. Indeed, similar molecules may have a different activity profile for the same array of properties, leading to a different clinical profile. Minor changes are sometimes responsible for the difference between “launched” and “withdrawn”. The “Black Swan”, a metaphor for the impact of the highly improbable, appears to be an inherent quality of the drug discovery enterprise. Predictions are based on the past, and can not substitute for proper design and use of experiments. Systems chemical biology may alleviate some of the inherent limitations related to prediction, by better integrating available experimental data.