I study the typical configuration of a C_p plot when the number of variables in the regression problem is large and there are many weak effects. I show that a particular configuration that is very commonly seen can arise in a simple way. I give a formula by means of which the risk incurred by the “minimum C_P ” rule can be estimated.