The use and interpretation of results from canonical correlation analysis have always been difficult. This paper attempts to resolve some of the problems involved by establishing the connection between canonical correlation and multiple regression on one hand, and canonical correlation and simultaneous equation systems on the other.