Assume that (X_n，?_n)^∞₁ is an adapted reward sequence， (γ_n）^∞₁) is snell's envelope of X_n，(A_n) is a predictable increased process with A₀=0 in Doob-Meyer's decomposition of (γ_n). In this paper，from J. Klass，we prove that σ₁=inf{K≥1:Xk₁≥γ_k} is the smallest semi-optimal and the largest strong regular generalized rule，and K₀?sup{n≥0:A_n=0}<∞ is the largest reqular generalized rule. Hence another expression of the uniqueness of optimal generalized rule is given.