The development of granular carbon materials with outstanding selectivity for the separation of alkenes and alkanes is highly desirable in the petrochemical industry but remains a significant challenge due to closely similar molecular sizes and physical properties of adsorbates. Herein, we report a facile approach of using natural biomass to prepare novel granular carbon molecule sieves with a molecular recognition accuracy of 0.44 Å and propose a new three-region model for the pore size distribution of amorphous porous carbons. Coffee bean-based granule carbon molecular sieves (CFGCs) were prepared with precise micropore regulation with subangstrom accuracy and characterized using molecular probes to reveal the evolution of carbon structure during preparation. The CFGC-0.09-750 demonstrates exceptional selectivity adsorption toward C3H6 while excluding C3H8, with an uptake ratio of 106.75 and a C3H6 uptake of 1.88 mmol/g at 298 K and 100 kPa, showcasing its immense potential in industrial applications for separating C3H6 and C3H8. The novel three-region model established in this work can clearly and reasonably elucidate why the samples CFGCs can screen propylene from propane at the subangstrom level. This study provides important guidance for the development of new carbon molecular sieves with subangstrom accuracy in molecular recognition and separation capacity.