Some difficulties with the likelihood principle, as formulated by A. Birnbaum, are discussed by means of an example that focuses attention on the deeper structure underlying conventional statistical models.A restricted version of the likelihood principle, compatible with the subjective Bayesian approach, is stated.