Objective: To examine statistical models that have been used to predict the cessation of breast-feeding. Setting: In nutritional epidemiology, a knowledge of risk factors that lead to breastfeeding cessation is essential to promote optimal infant health by increasing or sustaining breast-feeding rates. However, a number of methodological issues complicate the measurement of such risk factors. It is important when building multivariate models that variables entered into the model are not intervening variables, factors on the causal pathway or surrogate outcomes. Inclusion of these types of variable can lead to inaccurate models and biased results. A factor often cited to predict breast-feeding is 'intention to breast-feed' prior to the birth of the infant, although this factor is directly on the causal decision-making pathway. Another factor often cited is the age of introduction of formula feeding, which is actually part of the outcome variable because formula feeding defines the difference between full, complementary and no breast-feeding. Rather than include these as risk factors in multivariate models, factors removed from the causal pathway such as influences of educational practices, including advice to complementary feed, and beliefs and attitudes of families and health-care practitioners should be measured. Conclusions: The accurate quantification of modifiable risk factors is essential for designing public health education campaigns that are effective in sustaining or increasing breast-feeding duration.Keywords Breast-feeding epidemiology Breast-feeding promotion Infant health Risk factors Statistical methodsIn this paper, we use the following definitions:. Full breast-feeding to describe either exclusive breastfeeding, i.e. infants who receive breast milk as their only source of nutrition except for vitamin/mineral supplements or medicines, or predominant breast-feeding, i.e. infants who are breast-fed and receive liquids (water and water-based drinks, fruit juice, oral rehydration solution), ritual fluids and vitamin/mineral supplements or medicines, but who do not receive infant formula or solid or semi-solid foods. . Complementary breast-feeding to describe infants who receive breast milk but who also receive infant formula or solid or semi-solid foods. . No breast-feeding to describe infants who receive no breast milk.A large body of evidence for risk factors that are associated with the initiation and duration of breast-feeding has been reviewed 1 . However, a number of key issues complicate the interpretation of risk factor evidence. One problem that sometimes arises is that methodologically sound approaches to building multivariate statistical models are not always used. An issue that is frequently overlooked is that any variable considered in a model must be a potentially predictive factor but must not be an intervening variable, a factor on the causal pathway or a surrogate outcome 2,3 . The inclusion of these types of variable in multivariate models can bias measures of effect and mask...