Source scaling relation is studied over the magnitude (mN) range 2.6 to 5.4 using P waves generated by 31 tightly clustered earthquakes in New Brunswick, Canada. The recording stations, six in total, have a 60-points/sec sampling rate and a dynamic range of about 100 dB. They are located at regional distance (188 to 448 km), with a wide azimuthal coverage. The data interpreted consist of 115 spectral ratio curves (2-20 Hz), each obtained in a manner that allows effective cancellation of the effects caused by source radiation pattern, path attenuation, geometrical spreading, instrument error, and variability in site function. The data selected in this study differ from the single-station records used in a previous source-scaling study of Miramichi earthquakes (Chael, 1987) in having: 1) broader distance coverage; 2) greater recording dynamic range; 3) higher Nyquist frequency; and 4) larger data size. We conclude from the observed spectral ratios that source models having an ω−2 high-frequency fall-off (ω-square model) are strongly favored by the data over those having an ω−3 high-frequency fall-off (ω-cube model) and that stress drop increases with moment at a rate proposed earlier by Nuttli (1983a, b).