We examine the scaling of spectral properties of 122 local earthquakes (2.6<ML<5.5) recorded by a network of 10 broadband seismic stations during the period January 2001–November 2007 in northeast India. The spectral (low‐frequency plateau and corner frequency) and source (static stress drop, radiated energy, and radiation efficiency) parameters are obtained assuming an omega‐square source model. The source parameters are obtained from the displacement spectra corrected for frequency‐dependent attenuation and site effects, and limited frequency bandwidth. The well‐resolved estimates of these parameters, corresponding to corner frequencies within the maximum frequency resolution limit, are included in the analysis. We discuss the self‐similar scaling by introducing a nondimensional parameter that is proportional to stress drop independent of any particular fault geometry and by considering the measurement bias. For the events used in this study, the value of Cr is close to 2.0, a value expected for circular crack model and self‐similar source scaling. However, the relatively large uncertainties due to earthquake selection bias and recording bandwidth limitation suggest caution in drawing conclusions from these findings. The events are grossly characterized by low and almost constant value of radiation efficiency and positive overshoot corresponding to an increase in relative dynamic strength, suggesting the importance of energy dissipation during the rupture process. Although the scaling of radiated energy and stress drop, as obtained here, takes into account as much as possible the effects of finite bandwidth, attenuation, and site corrections, the limited event selection may strongly bias it. An improvement over the present scaling is suggested with an improved seismic network in the region and also with the use of a stable empirical Green’s function approach for source parameter estimation.