In this work, we present a simple approach for devising order parameters (OPs) for atomic systems based on pattern recognition techniques. It exploits the fact that all crystalline substances are characterized by a unique "signature" cell (SC) which is constructed using a central atom and its nearest N-SC neighbors in a given crystal. The algorithm measures the local degree of similarity between a SC and the system to be analyzed. The best fit of a SC to N-SC atoms surrounding a given atom in the system is determined by maximizing a fictitious energy of binding among those atoms and the SC atoms. The fictitious potential energy is designed to give maximum attractive energy for maximum overlap. The maximum binding energy of interaction attained in this process is used as a measure of similarity between the crystal structure and the system (i.e., as an OP). The proposed method provides a unified and intuitive approach for constructing relevant OPs for a given system. We used these OPs to characterize the order of different phases in the Lennard-Jones system and in a model silicon system. It is shown that these novel OPs give a more complete description and a better understanding of the structural order in amorphous silicon than conventional OPs.