Despite the growing interest in dietary patterns, there have been few longitudinal investigations. The objective of the present study was to extend an earlier method of dietary pattern assessment to longitudinal binary data and to assess changes in patterns over time and in relation to socio-demographic covariates. A prospective national cohort of 1265 participants completed a 5 d food diary at three time-points during their adult life (at age 36 years in 1982, 43 years in 1989 and 53 years in 1999). Factor analysis identified three dietary patterns for women (fruit, vegetables and dairy; ethnic foods and alcohol; meat, potatoes and sweet foods) and two patterns in men (ethnic foods and alcohol; mixed). Trends in dietary pattern scores were calculated using random effects models. Marked changes were found in scores for all patterns between 1989 and 1999, with only the meat, potatoes and sweet foods pattern in women recording a decline. In a multiple variable model that included the three time-points, socio-demographic variables and BMI time-dependent covariates, both non-manual social class and higher education level were also strongly associated with the consumption of more items from the ethnic foods and alcohol pattern and the mixed pattern for men (P < 0.0001) and the fruit, vegetables and dairy pattern and the ethnic foods and alcohol pattern for women (P < 0.01). In conclusion, longitudinal changes in dietary patterns and across socio-economic groups can assist with targeting public health initiatives by identifying stages during adult life when interventions to improve diet would be most beneficial to health.