We study the problem of synthesizing regular expressions from a set of positive and negative strings. The previous synthesis algorithm proposed by Lee et al. [12] relies on the best-first enumeration of regular expressions. To improve the performance of the enumeration process, we define a new normal form of regular expressions called the concise normal form which allows us to significantly reduce the search space by pruning those not in the normal form while still capturing the whole class of regular languages. We conduct experiments with two benchmark datasets and demonstrate that our synthesis algorithm based on the proposed normal form outperforms the previous algorithm in terms of runtime complexity and scalability.