It's well known that the schemata theorem and the implicit parallelism are two basic theoretical foundations of genetic algorithms (GA). In this paper, the authors analyze the two basic principles and show that the two principles are not strict and have some disadvantages. That is, as the bases of GAs, the theorems are not perfect. In order to deepen the comprehension of GA, a new ideal density model of GA is presented in this paper. Based on the model, it's known that the GA is actually a guiding stochastic search. And the searching direction is guided onto the chromosome family whose ancestors belong to schemata with high fitness. Using the model to solve the typical function optimization problem, the simulation results show that the new GA has much better speed and can get more precise results. This shows that the new GA model has potential usage in practice.