In different contexts protein function can mean very different things: it may refer to the nitty-gritty of the protein's biochemical activity; it can mean the cellular process in which the protein is involved; it can refer to its physiological role or to its ostensible connection to phenotype or disease. Indeed, function prediction methods typically focus on a very specific aspect of function. However, the different facets of protein function are tightly related: understanding the biochemical function sheds light on the physiological function and may help explain the effect of the on phenotype. I will show that systems level analysis can unravel the molecular mechanisms of individual proteins, and that understanding the molecular details of individual proteins can illuminate physiology beyond the cellular level. Furthermore, I will describe a multilevel approach for function prediction that combines molecular data with cellular, physiological, clinical and network data to promote function prediction at all levels.