With so many spectroscopic surveys, both past and upcoming, such as SDSS and LAMOST, the number of accessible stellar spectra is continuously increasing. There is therefore a great need for automated procedures that will derive estimates of stellar parameters. Working with spectra from SDSS and LAMOST, we put forward a hybrid approach of Kernel Principal Component Analysis (KPCA) and Support Vector Machine (SVM) to determine the stellar atmospheric parameters effective temperature, surface gravity and metallicity. For stars with both APOGEE and LAMOST spectra, we adopt the LAMOST spectra and APOGEE parameters, and then use KPCA to reduce dimensionality and SVM to measure parameters. Our method provides reliable and precise results; for example, the standard deviation of effective temperature, surface gravity and metallicity for the test sample come to approximately 47–75 K, 0.11–0.15 dex and 0.06–0.075 dex, respectively. The impact of the signal:noise ratio of the observations upon the accuracy of the results is also investigated.