A few approaches for handling baseline responses are available for use in pharmacokinetic (PK)-pharmacodynamic (PD) analysis. They include: (method 1-B1) estimation of the typical value and interindividual variability (IIV) of baseline in the population, (B2) inclusion of the observed baseline response as a covariate acknowledging the residual variability, (B3) a more general version of B2 as it also takes the IIV of the baseline in the population into account, and (B4) normalization of all observations by the baseline value. The aim of this study was to investigate the relative performance of B1-B4. PD responses over a single dosing interval were simulated from an indirect response model in which a drug acts through stimulation or inhibition of the response according to an Emax model. The performance of B1-B4 was investigated under 22 designs, each containing 100 datasets. NONMEM VI beta was used to estimate model parameters with the FO and the FOCE method. The mean error (ME, %) and root mean squared error (RMSE, %) of the population parameter estimates were computed and used as an indicator of bias and imprecision. Absolute ME (|ME|) and RMSE from all methods were ranked within the same design, the lower the rank value the better method performance. Average rank of each method from all designs was reported. The results showed that with B1 and FOCE, the average of |ME| and RMSE across all typical individual parameters and all conditions was 5.9 and 31.8%. The average rank of |ME| for B1, B2, B3, and B4 was 3.7, 3.8, 3.3, and 5.2 for the FOCE method, and 4.6, 4.3, 4.7, and 6.4 for the FO method. The smallest imprecision was noted with the use of B1 (rank of 3.1 for FO, and 2.9 for FOCE) and increased, in order, with B3 (3.9-FO and 3.6-FOCE), B2 (4.8-FO; 4.7-FOCE), and B4 (6.4-FO; 6.5-FOCE). We conclude that when considering both bias and imprecision B1 was slightly better than B3 which in turn was better than B2. Differences between these methods were small. B4 was clearly inferior. The FOCE method led to a smaller bias, but no marked reduction in imprecision of parameter estimates compared to the FO method.