Genetic Algorithm, a nonlinear parallel algorithm, is being applied in more and more fields. In this paper, two kinds of new mutation technique are introduced, and fitness function modification is executed depending on the difference between maximal fitness and minimal fitness, three operators (crossover, replication, mutation) are rearranged to enlarge search range and sorting is changed with the searching process according to the repeating number of some individuals in population. Finally, a typical template class of genetic algorithms , which can be used to handle the integer or float type sequence with any number of variables, has been built in Visual C 4.2 development environment. Based on it, (1) any derivative classes including users' special fitness function can be written easily. Some function, such as stop criteria function and fitness modification function can be rewritten for users' specification. Testified by classical test functions, these improvement (2) shows good search efficiency and not easily fall into sub-optimal solution like traditional search methods do.