Marginal and joint nonlinear principal components are defined for normally distributed random vectors. Their properties are investigated. Next, the proposed approach is compared with some generalisations of principal components analysis known in literature.