Reactive systems are. generally, highly successful for dynamic uncertain domains. Analysing why this should be so indicates the main criterion for success is that information apprehended about a given situation can render it certain enough to reliably inform action. A case is made for informing action by dynamic world modelling for domains where this does not hold. When the issue of attending selectively to situational information is introduced, the case for dynamic world modelling is enhanced. A system incorporal~g selective perception which dynamically models its domain is described.