A broad class of theories exist which share the distinguishing characteristics of quantum mechanics, but allow even stronger correlations. Therefore a criterion that could be used to distinguish physical theories from non-physical ones would be of considerable value. The principle of 'information causality', introduced here by Marcin Pawłowski et al., may provide just this. The principle states that communication of m classical bits causes information gain of at most m bits. The authors show that it is respected by classical and quantum physics, but violated by other models that resemble quantum mechanics but with stronger correlations. The principle is a generalization of the no-signalling condition (information cannot be transmitted faster than light) and may be a foundational property of nature. A broad class of theories exist which share the distinguishing characteristics of quantum mechanics but allow even stronger correlations. Here, the principle of 'information causality' is introduced and shown to be respected by both classical and quantum physics; however, it is violated by other models that resemble quantum mechanics but with stronger correlations. It is suggested that information causality may help to distinguish physical theories from non-physical ones. Quantum physics has remarkable distinguishing characteristics. For example, it gives only probabilistic predictions (non-determinism) and does not allow copying of unknown states (no-cloning1). Quantum correlations may be stronger than any classical ones2, but information cannot be transmitted faster than light (no-signalling). However, these features do not uniquely define quantum physics. A broad class of theories exist that share such traits and allow even stronger (than quantum) correlations3. Here we introduce the principle of ‘information causality’ and show that it is respected by classical and quantum physics but violated by all no-signalling theories with stronger than (the strongest) quantum correlations. The principle relates to the amount of information that an observer (Bob) can gain about a data set belonging to another observer (Alice), the contents of which are completely unknown to him. Using all his local resources (which may be correlated with her resources) and allowing classical communication from her, the amount of information that Bob can recover is bounded by the information volume (m) of the communication. Namely, if Alice communicates m bits to Bob, the total information obtainable by Bob cannot be greater than m. For m = 0, information causality reduces to the standard no-signalling principle. However, no-signalling theories with maximally strong correlations would allow Bob access to all the data in any m-bit subset of the whole data set held by Alice. If only one bit is sent by Alice (m = 1), this is tantamount to Bob’s being able to access the value of any single bit of Alice’s data (but not all of them). Information causality may therefore help to distinguish physical theories from non-physical ones. We suggest that information causality—a generalization of the no-signalling condition—might be one of the foundational properties of nature.