Historically, the ability of foods to support the growth of spoilage organisms and food-borne pathogens has been assessed by inoculating a food with an organism of interest, and following its growth over a period of time. Information gained from such challenge tests, together with knowledge of the organoleptic stability of the product, can then be used to determine an appropriate shelf-life for the food. Whilst this approach may be seen as the "gold-standard" of microbiological assessment of food, it is both time-consuming and costly. A major advance to complement challenge testing was the development of predictive modelling, when it was demonstrated that the growth of a wide range of organisms of interest could be quite accurately modelled as a function of only a few environmental parameters-primarily temperature, pH and water activity (a(w)), with perhaps other factors such as nitrite, organic acids and oxygen. This approach to predictive microbiology is embodied in software tools such as the UK Food MicroModel and the Pathogen Modeling Program from the USA. Whilst modelling of this form yields accurate predictions of the growth of organisms in the majority of foods, there are occasions when there are discrepancies between the model and the observed growth. These discrepancies are most often described as "fail-safe", i.e. the observed growth is slower than predicted by the model. This paper examines the role of food structure in the development of microbial populations and communities, and describes the methodologies we propose to begin to tackle some of these complex and interlinked issues.