We have derived an empirical attenuation curve for a local magnitude scale in northwest Iran using a large data set of local earthquakes that occurred from 1996 to 2009. Investigation shows that using one-half the peak-to-peak value tends to underestimate the magnitude of an event by as much as 0.07 magnitude units in comparison with using the zero-to-peak value. Additionally, using a vector sum of horizontal measurements overestimates the ML values by 0.16 magnitude units in comparison with magnitude values that are determined using arithmetic means of horizontal measurements. We simulated the corresponding Wood–Anderson seismograms for over 133,000 available waveforms. After preliminary analysis, the data set was subdivided in order to derive a local magnitude scale for northwestern Iran. The distance attenuation curve, station correction terms, and the magnitude of events were simultaneously estimated using parametric and nonparametric approaches. In both approaches, estimations and separate attenuation curves were determined by using vertical and horizontal measurements. These two approaches yielded very similar results. The distance attenuation curves that resulted in a nonparametric approach are given by -logA0=0.9252log(Δ)+0.0030Δ+0.8496 and -logA0=0.9993log(Δ)+0.0029Δ+0.7114 for the vertical and horizontal components, respectively, where Δ is the epicentral distance in kilometers. The distances considered ranged from 1 to 500 km. Applying these distance-correction terms to the data set show that there are no obvious trends in magnitude residual with distance. Also, using the station corrections reduces the magnitude residuals by as much as 0.07 magnitude units.