Numerous epidemiological studies of various kinds of stress and preterm delivery have produced either negative or weakly positive results. Those inconclusive findings could be either because of the absence of an association or because of recognised methodological problems that may have masked an association. The biological plausibility of the stress hypothesis provides one rationale for continuing stress research, using better study designs. To further this agenda, we propose an epidemiological model, based on the classic "host, environment, agent" triangle of epidemiological causality. The host is the individual woman, more or less susceptible to stressor-induced pathology. The environment includes the social and cultural conditions that are ongoing stressors as well as social and cultural modifiers of stress e.g. those factors that may influence how a particular stressor is experienced or what the physical response to it may be. The agent is the immediate emotional or physical stressor requiring her response. We draw from recent literature, published principally since 1990, to illustrate this model. This epidemiological model posits that whether the individual is overwhelmed by stressors depends not only on the strength of the agents but also upon host susceptibility to stress, as well as the background level of acute, environmental and contextual stressors, and the moderating influence of host, environmental and contextual resources for handling stress. Future research needs to be based on stress hypotheses that include all sides of the triangle, data collection instruments that adequately capture relevant stressors and stress responses, and analytical techniques capable of handling complex, multilevel relationships.