This paper deals with the estimation of the moderator temperature coefficient of reactivity (MTC) by noise analysis. The current noise analysis-based MTC measurement, relying on the cross correlation between the neutron noise measured by a single in-core neutron detector and the local temperature noise given by a single core-exit thermocouple located at the top of the same fuel assembly, or of a neighboring fuel assembly, is not accurate. The MTC is systematically underestimated by a factor of 2 to 5 compared to its design-predicted value. A theoretical study shows that, in case of nonhomogeneous moderator temperature noise, the core-averaged moderator temperature noise should be used for the MTC estimation. The new estimation method can reach up to 3% accuracy as compared with the results of core calculations for the Swedish Ringhals-2 pressurized water reactor (PWR). We show via noise measurements performed at the Ringhals-2 PWR that the moderator temperature noise is actually radially strongly heterogeneous and loosely coupled. The new MTC noise estimator is demonstrated to provide an accurate MTC evaluation, with the core-averaged moderator temperature noise estimated via the use of many radial in-core gamma-thermometers. More important, different forms of weighting functions are suggested to calculate the core-averaged moderator temperature noise. This new MTC noise estimator, which is nonintrusive and free of calibration, can therefore be applied to monitor the MTC throughout the cycle.