Background: Antibiotic-impregnated calcium sulfate delivery systems (ACDS) are commonly used to treat chronic osteomyelitis. Our research is to investigate drug release in vitro over a longer period, as a cautious predictor of in vivo release. Methods: The local release behavior of antibiotic in vitro was simulated. The consecutive dynamic eluting experiment was performed based on the pro-operative characteristic of osteomyelitis patients and the determined results of drug concentration in the human drainage tissue fluid (DTF). The concentration of each drug in the receiving solution was detected by ultra-performance liquid chromatography-tandem quadrupole detector mass spectrometry. The ACDS was reviewed by scanning electronic microscopy (SEM) after 48 h, and prepared to be eluted for another examination after 33 days. The mechanism of antibiotic release was analyzed by using the Ritger–Peppas and Weibull equations. Results: The cumulative release rate of vancomycin in a vancomycin-calcium sulfate delivery system (VCDS) was 77.50 % (3.0 mm diameter) and 72.43 % (4.8 mm diameter), while that of the tobramycin in a tobramycin-calcium sulfate delivery system (TCDS) was 88.0 % (3.0 mm diameter) and 84.55 % (4.8 mm diameter). At the 15th day, approximately 27.92% of vancomycin was and 29.35% of tobramycin was released from the local implant in vivo. Using SEM, numerous vancomycin and tobramycin particles were found to be attached to the columnar calcium sulfate crystals at the start of the experiment. The release behavior of the two antibiotics followed a combination of Fickian diffusion and Case II transport mechanisms within the first 48 h, and a Fickian diffusion mechanism during the subsequent time period. The correlation coefficient of tobramycin and vancomycin in vivo and in vitro was 0.9704–0.9949 and 0.9549–0.9782, respectively. Conclusion: A good correlation of the in vivo and in vitro cumulative release rates was observed by comparing the cumulative release rate of drugs in vitro by means of the dynamic eluting model, and in the DTF. Therefore, our study has proved that it is possible to use the dynamic eluting model as a cautious predictor of in vivo release.