The inaccuracy of the Black-Scholes formula arises from two aspects: the formula is for European options while most real option contracts are American; the formula is based on the assumption that underlying asset prices follow a lognormal distribution while in the real world asset prices cannot be described well by a lognormal distribution. We develop an American option pricing model that allows non-normality. The theoretical basis of the model is Gaussian quadrature and dynamic programming. The usual binomial and trinomial models are special cases. We use the Jarrow-Rudd formula and the relaxed binomial and trinomial tree models to imply the parameters related to the higher moments. The results demonstrate that using implied parameters related to the higher moments is more accurate than the restricted binomial and trinomial models that are commonly used.