In this study, we focus on the steel plate cutting problem (SPCP), where a set of rectangular order plates with specified demand are cut from large rectangular steel plates. The aim of solving SPCP is to minimize the number of steel plates used in the cutting process. According to the analysis of the cutting production line in steel mills, we regard the SPCP as a two-dimensional non-exact three-stage cutting stock problem (2D3SCSP) with two additional practical constraints. Since these two practical constraints limit the length between any two adjacent guillotine cuts in the first stage by a predetermined parameter and the number of guillotine cuts in the second stage by one, the existing solving methods proposed for 2D3SCSP cannot be used for SPCP directly. Four heuristics based on column generation (HCG) are proposed to solve SPCP. The performance of the four HCGs is analyzed through conducting a set of experiments, and an effective HCG with the ability of obtaining high-quality solutions within acceptable computational times is finally obtained.