The centers method (Cazes et al., 1997, Chouakria, 1998) was the first principal component analysis for interval-valued data (where it is implicitly assumed that values within an interval are uniformly distributed across that interval). Many other methods have since been proposed. All fail in various ways to capture fully all the information contained in the data. Here, we set these in context against a new method which calculates the covariance matrix exactly. This new method also includes a new visualization of the projection of the observations onto the principal component space.