A methodology for autotuning decentralized proportional-integral-derivative (PID) controllers for multivariable systems is proposed. It combines the autotuning procedure involving sequential loop closing and relay tests with time-domain curve fitting via least squares to identify suitable transfer function models that account for interactions among the loops, using limited response data. Simulation results reveal that the curve fitting is able to identify models using only the first few oscillations of the relay-test response. Thus, there is no need to wait for the process to reach steady oscillations. A controller tuning method developed by Semino and Scali (J. Process Control 1998, 8 (3), 219-227) is adopted to tune the PI controllers. Detailed results on several 2 × 2, 3 × 3, and 4 × 4 multivariable processes show that the proposed methodology gives PI controllers having performances comparable with those tuned by the ATV+ method of Parabita et al. (Sequential Identification and Autotuning by Relay Techniques of Decentralised Controllers for MIMO Processes, ADCHEM 2000, Pisa, Italy, 2000 or Luyben's BLT technique (Ind. Eng. Chem. Process Des. Dev. 1986, 25 (3), 654-660). The major advantage of the proposed autotuning with least-squares is the significant reduction in the duration of relay tests. Computations involved in the methodology can be carried out on a personal computer in reasonable time.