Abstract The Mixed-Integer Linear Programming models are a common representation of real-world objects. They support simulation within the expressed bounds using constraints and optimization of an objective function. Unfortunately, handcrafting a model that aligns well with reality is time-consuming and error-prone. In this work, we propose a Grammatical Evolution for Constraint Synthesis (GECS) algorithm that helps human experts by synthesizing constraints for Mixed-Integer Linear Programming models. Given relatively easy-to-provide data of available variables and parameters, and examples of feasible solutions, GECS produces a well-formed Mixed-Integer Linear Programming model in the ZIMPL modeling language. GECS outperforms several previous algorithms, copes well with tens of variables, and seems to be resistant to the curse of dimensionality.