Cells have developed complex control networks which allow them to sense and response to changes in their environment. Although they have different underlying biochemical mechanisms, signal transduction units in prokaryotes and eukaryotes fulfill similar tasks, such as switching on or off a required process or amplifying a certain signal. The growing amount of data available allows the development of increasingly complex models which offer a detailed picture of signaling networks, but the properties of these systems as a whole become difficult to grasp. A sound strategy to untangle this complexity is a decomposition into smaller units or modules. How modules should be delimited, however, remains an unanswered question. We propose that units without retroactive effects might be an interesting criterion. In this contribution, this issue will be explored through several examples, starting with a simple two-component system in Escherichia coli up to the complex epidermal growth factor signaling pathway in human cells. (c) 2004 Elsevier Ltd. All rights reserved.