<p>A new non-parametric approach to derive the principal components (PCs) of interval-valued data with multiple observations from the same subject using patterned covariance structures is proposed. We exploit the patterned covariance structures to take into account between subject variation, within interval variation and within multiple observations (sequences or brands) variation of the interval-valued data in the PCs. This is done in two stages: first getting eigenblocks and eigenmatrices of the patterned variance-covariance matrix, and then analyzing these eigenblocks and the corresponding principal vectors together to get the PCs of the interval-valued data. To the best of our knowledge this is the first study that takes into account both within interval variation and within sequence/brand variation along with the between subject variation to derive the PCs and detect any sequence or brand effect. The proposed method is efficient in analyzing interval-valued datasets with multiple observations from each subject. Results illustrating the accuracy and appropriateness of our new method over the existing methods are presented. Our proposed exploratory method is computationally efficient for large multivariate interval-valued datasets with multiple number of observations from each subject and is illustrated with two real data examples.</p>