Type classification of Holstein cows has been done in Canada since 1925 by the breed association. Scoring is done once in life, but cows are reclassified only if their score can go upwards. Type breeding values are estimated using a single trait animal model which fails to account for the relationship between age and type. Random regression models have been advocated to analyze longitudinal data because they allow for individual deviations from an average curve. Therefore, animals that improve or deteriorate their type genetic merit with age can be identified. An American Holstein data set, containing information on 655,899 cows from 6,245 herds in 7 states, was used to implement a random regression model on 8 type traits. Stature, rump angle, thurl width, rear leg set, rear udder width, rear udder height, udder depth, and fore udder attachment were analyzed separately using a single trait, random regression animal model. This included the fixed effects of: age and age squared at classification within year of birth, herd-year-classifier, and months in milk. The model also included the random effect of permanent environment, and a random animal effect which was modelled as an intercept, age and age squared at classification. Variance components were estimated using Bayesian theory. Samples from marginal posterior distributions of the parameters of interest were obtained utilizing Gibbs sampling. Breeding values were also predicted using a single record (first lactation record), single trait animal model similar to the current Canadian conformation model, and a repeatability model using all records. Breeding values from the random regression model, at different ages, were correlated to those computed by the Canadian and repeatability models. Predicted transmitting ability for productive herd life of sires with 100 or more daughters (955) were correlated with estimated breeding values from the three models. The random regression model was able to detect individual changes in breeding values over time. Genetic variances in all traits increased with age. Genetic correlations among ages within traits were all positive. Higher genetic correlations were observed between ages that were closer together in time. On average, across traits, 12.9% of the sires with 100 or more daughters decreased their breeding value as age progressed. Spearman correlations among breeding values from the Canadian and random regression models were low to moderate, at young ages, and low at later ages. Correlations were higher among breeding values from the repeatability and random regression models. Results of this study showed that genetic variation over time exists in type traits which can be exploited in conformation trait genetic programs. Due to low rank correlations at older ages, prediction of type additive genetic merit at mature ages based on results of the single trait Canadian model is not recommended. Estimated breeding values for mammary traits from the three models were positive and had moderately low correlations to herd life. Rear leg set and thurl width were negatively correlated to herd life. This indicated that cows with more curved legs and/or wider rump had shorter lives. Stature and rump angle were independent of herd life. Correlations between herd life and breeding value for type from the random regression model decreased with age. Breeding values from the random regression model did not predict herd life better than breeding values from the Canadian and repeatability models.