This chapter looks at the correlation between two arrays. If we have two arrays, each containing the values for its own set of variables on the same set of individuals, a legitimate question is whether or not the information provided by these two sets of data is redundant. One answer is provided by Canonical Correlation Analysis, developed in this chapter. The basic idea is to extend the notion of correlation between two variables to a correlation between two arrays. The two sets of variables are considered to be correlated if each variable in one array is a linear combination of the variables in the other array. We will therefore look for axes for each of the two arrays such that the components associated with them are the most correlated. A geometric solution using projectors is proposed. However, while this result is very intuitive, it is not optimal for the complexity of the calculations. The proposed algorithm for calculating the canonical axes and components is based on a reorganization of the calculations to reduce them to a PCA with metrics of the cross-tabulated array of the two initial arrays. This approach will be used in the next chapter to extend the canonical analysis to more than two tables.