Metabolic pathway design is an essential step in the course of constructing an efficient microbial cell factory to produce high value-added chemicals. Meanwhile, the computational design of biologically meaningful metabolic pathways has been attracting much attention to produce natural and non-natural products. However, there has been a lack of effective methods to perform metabolic network reduction automatically. In addition, comprehensive evaluation indexes for metabolic pathway are still relatively scarce. Here, we define a novel uniform similarity to calculate the main substrate-product pairs of known biochemical reactions, and develop further an efficient metabolic pathway design tool named PyMiner. As a result, the redundant information of general metabolic network (GMN) is eliminated, and the number of substrate-product pairs is shown to decrease by 81.62% on average. Considering that the nodes in the extracted metabolic network (EMN) constructed in this work is large in scale but imbalanced in distribution, we establish a conditional search strategy (CSS) that cuts search time in 90.6% cases. Compared with state-of-the-art methods, PyMiner shows obvious advantages and demonstrates equivalent or better performance on 95% cases of experimentally verified pathways. Consequently, PyMiner is a practical and effective tool for metabolic pathway design.