Regulatory guidelines define a highly variable drug (HVD) as a drug exhibiting >30% within-subject variability (WV) in terms of the area under the plasma concentration-time curve (AUCt) and maximum plasma concentration (Cmax) parameters. However, it can be difficult to identify an HVD because clinical trials seeking to define WVs are complex. We explored whether population PK analysis method reliably estimated residual variability (RV) compared to WV, and confirmed that such approaches identified HVDs using real-world data (RWD). We simulated 1,000 datasets based on a one-compartment intravenous model with 12 sampling points for each of 6 ,12, 18, 24, or 30 subjects; five levels of WV (10, 20, 30, 40, and 50%) in terms of pharmacokinetic (PK) parameter distribution; either without inter-individual variability (IIV) (0%) or with IIVs of 10, 20, 30, 40, and 50. The datasets were generated using R (Ver 3.6.0) assisted by R Studio(Ver. 1.2.1335) software and population analysis was performed using NONMEM ver. 7.4.0 software. We used a proportional error model and the first-order conditional estimation method with the interaction option (FOCE-I). After PK modeling, we determined the proportion of sigma values within ±10% of the true values. Earlier RWD 3x3 bioequivalence data of a well-known HVD, were replicated using our method. When the IIV was zero, the success rate of NONMEM estimation was >90% for WVs from 10 to 30% and subject numbers >12. At a WV of 40%, the success rate was 90% when the subject number was >18. When the IIV changed from 10 to 50%, the success rate was >90% for WVs from 10 to 40% and subject numbers >18. However, the residual variability (RV) estimated by NONMEM underestimated the WV when the WV was 50%. In the RWD applications, for the eperisone case, the RVs were 40.9, 45.8, 42.7, 43.8, and 45.1% when data were collected from 6, 12, 18, 24, and 30 random subjects, they were reliable results referenced by the within-subject coefficient of variation (CVwR) of eperisone for the Cmax (50.21%). As the result of the simvastatin case, the RVs were 38.2, 46, and 44.4% for each scenario of the 6, 12, and 22 subjects, they were approximated to result referenced by the CV of simvastatin for Cmax (37.14%). In the risperidone case, the RVs were 26.3, 19.1, and 23.2% for each scenario of 6, 12, and 23 subjects, they were close with the result cited by the CVwR of risperidone for Cmax (31.5%). Based on the results of this study, we confirmed that the population PK analysis method well-estimated WVs from population PK data. Hence, we suggest that it can be used to identify HVDs.