The Mw 6.3 Central Italy (L'Aquila) earthquake on April 6, 2009 has been recorded by the Irpinia Seismic Network (ISNet) about 250 km SE of the epicenter. The ISNet array has an aperture of about 80 km and consists of 25 stations, each equipped with a 3C CMG-5T accelerometer. Waveforms from 19 stations could be used to estimate rupture geometry, event magnitude, and moment tensor. The recorded, low-pass filtered waveforms (f<2 Hz) are very coherent and permit the application of array methods to measure backazimuth and slowness of the incoming waves for subsequent event location. To image the rupture geometry we implemented a modified beamforming and stacking technique that back-projects the recorded direct P-wave amplitudes into the earthquake source region using travel times from a 1D velocity model. Amplitudes were measured in a 0.5 s moving time window until the arrival of secondary phases. A NW-SE striking rupture of 17 km length is imaged, propagating with an average velocity up to 3 km/s. The P- and S-wave displacement spectra recorded by the 3C accelerometric and some broad-band sensors have been inverted to determine the main source parameters. We assumed an omega-square spectral model and a constant-Q attenuation function, with a parameter t* (T/Q) directly estimated from the high-frequency spectral decay. The estimated seismic moment (2.1 x 10^18 Nm) is consistent for P- and S-waves within a factor 2, providing a maximum Mw 6.1. This value is also consistent with the local magnitude inferred from Wood-Anderson synthetic records. Relatively high P- and S-corner frequencies (0.6 and 0.4 Hz), due to directivity effects, are the cause for an underestimation of the rupture length and an extremely high static stress drop value. On the other hand, an apparent stress of about 1.5 Mpa is measured from the radiated seismic energy, which is less sensitive to source directivity effects. We determine the moment tensor solution for the earthquake by modeling the strong-motion waveforms using two different approaches. The first one uses a point source approximation and a grid search over a set of trial source positions and time shifts to identify the optimal centroid position, time, and moment tensor. In the second method the rupture is represented by a finite 1D source model. Source finiteness is approximated by a summation over point sources aligned along the fault strike. The focal mechanism and the linear seismic moment distribution along the strike are inverted simultaneously by an optimized grid search combined with a simulated annealing algorithm. Moreover the technique provides some insight on the modality of the rupture. We find a centroid depth of about 5 km and a prevalently normal fault plane solution with a dominant directivity effect toward SE. Our studies demonstrate that the use of array techniques and a dense accelerometer network can provide quick and robust estimations of source parameters of moderate-size earthquakes located outside the network.