The paper presents a study of the critical flow phenomena modeling with MELCOR 2.2.15254 severe accident computer code. The Marviken Critical Flow Test number 21 (CFT-21) experiment was selected as a representative critical flow-focused Separate Effect Test (SET). Various modeling aspects were investigated, including the nodalization, model setup, parameters, and sensitivity coefficients. A local-type sensitivity study was performed to analytically identify the significant parameters and assess their impact on the modeling. A dedicated regression-based approach, using standard deviation, was developed to find the best-fit MELCOR modeling parameters. The primary purpose of this work was to determine the appropriate approach to model critical flow with MELCOR 2.2, investigate the model performance, assess the influence of nodalization choices, identify significant sensitivity parameters, and prepare recommendations with an emphasis on best-estimate modeling. An additional outcome was the benchmark of the recent code version with the Marviken test.