Research into the problem of splice site selection has followed a reductionist approach focused on how individual splice sites are recognized. Early applications of information theory uncovered an inconsistency. Human splice signals do not contain enough information to explain the observed fidelity of splicing. Here， we conclude that introns do not necessarily contain 'missing' information but rather may require definition from neighboring processing events. For example， there are known cases where an intronic mutation disrupts the splicing of not only the local intron but also adjacent introns. We present a genome-wide measurement of the order of splicing within human transcripts. The observed order of splicing cannot be explained by a simple kinetic model. Simulations reveal a bias toward a particular， transcript-specific order of intron removal in human genes. We validate an extreme class of intron that can only splice in a multi-intron context. Special categories of splicing such as exon circularization， first and last intron processing， alternative 5 and 3'ss usage and exon skipping are marked by distinct patterns of ordered intron removal. Excessive intronic length and silencer density tend to delay splicing. Shorter introns that contain enhancers splice early.