We estimate moment magnitudes Mw for earthquakes in Switzerland recorded between 1998 and 2009 using three different spectral methods. The Mw estimation in Switzerland is extended to lower magnitudes (local magnitude ML 0.1), and scaling relations between ML and Mw are investigated. Above ML 4, the obtained Mw estimates are consistent with the previously obtained scaling relation of Mw = ML - 0.3 at the Swiss Seismological Service (SED). Below ML 4, all three methods indicate that a 1:1-type relationship is inappropriate. Therefore, we propose a new piecewise empirical scaling relation for earthquakes in Switzerland. The scaling is linear below ML 2 and above ML 4. To obtain a smooth transition between the two linear scales we fit a quadratic relation in between (2 ≤ ML ≤ 4). This scaling relation is also consistent with Mw estimates from moment-tensor (MT) solutions based on broadband waveform fitting of local earthquakes with ML > 3.0. We have tested all three methods carefully to ensure that the observed break in scale at around M 3 cannot be attributed to bias in the Mw determination. However, we cannot determine with certainty from the dataset at hand whether the break in scaling is due to bias in the routine determination of ML or to physical properties of the source.