The incomplete probabilistic information can lead to imprecision. The traditional power system risk assessment can only deal with random information, but it cannot deal with the imprecise information. Interval Probability (IP) can reflect both randomness and imprecision. The risk assessment of power system based on IP is an effective method to deal with imprecision and can provide more useful information to the decision-makers. In the paper, the completeness degree of probabilistic information was depicted by IP. The nonlinear optimization models for imprecise risk assessment of generating system and composite generation-transmission system were established. And the genetic algorithm was used to obtain the upper and lower bound of reliability indices. Case study on revised IEEE-RTS79 and IEEE-RBTS system showed the rationality and equity of presented method.