The 2002–2003 Etna eruption is studied through earthquake distributions and surface fracturing. In September 2002, earthquake-induced surface rupture (sinistral offset ∼0.48 m) occurred along the E-W striking Pernicana Fault (PF), on the NE flank. In late October, a flank eruption accompanied further (∼0.77 m) surface rupturing, reaching a total sinistral offset of 1.25 m; the deformation then propagated for 18 km eastwards to the coastline (sinistral offset 0.03 m) and southwards, along the NW-SE striking Timpe (dextral offset 0.04 m) and, later, Trecastagni faults (dextral offset 0.035 m). Seismicity (<4 km bsl) on the E flank accompanied surface fracturing: fault plane solutions indicate an overall ESE-WNW extension direction, consistent with ESE slip of the E flank also revealed by ground fractures. A three-stage model of flank slip is proposed: inception (September earthquake), climax (accelerated slip and eruption) and propagation (E and S migration of the deformation).
 Flank instability affects numerous volcanoes worldwide, but its relationship with eruptive activity has so far remained poorly constrained. The 2002–2003 eruption of Mt. Etna (Italy) has permitted to study the initiation and growth of a major flank slip and its influence on volcanic activity. To evaluate the cause-effect relationships between flank slip and volcanism, we consider the distribution of seismicity and surface fracturing before and during the eruption.
 Seismicity at Etna is recorded by a permanent seismic network consisting of 30 stations with one-component short period (1 s) and three-component broad band (20 s) sensors. Earthquakes were located using the HYPOELLIPSE code [Lahr, 1989], considering the altitude of the stations and determining the hypocentre spatial patterns. The used 1D velocity model was derived by Hirn et al. . We selected 234 events (June 2002–February 2003; Figure 2), using the following thresholds: magnitude (Md) ≥2.5; horizontal 68% confidence limit in the least well-constrained direction (ERH) ≤1.0 km; 68% confidence limit for depth (ERZ) ≤1.5 km; root-mean-square residual (RMS) <0.4 s. The earthquakes used to compute fault plane solutions (FPSs) were characterized by high frequency content, sharp first arrivals and clear S-phases at the nearest stations, assuming therefore a double couple source. Analysis of FPSs was made using the FPFIT algorithm [Reasenberg and Oppenheimer, 1985] to plot first motion data and to evaluate nodal planes and orientation of P- and T-axes. Six FPSs relative to earthquakes with Md ≥3.0 (with number of polarities ≥14, number of polarity discrepancies ≤2, focal plane uncertainty ≤20°, unique and unambiguous solution) were obtained. Surface fractures (Figure 3) have been mainly measured (geometry, kinematics and displacement) along paved roads (error <5%) once or twice per week, starting September 2002.
3. Timing of Events: Earthquakes, Surface Fracturing and Eruption
 Seismicity at Etna during June–September 2002 was moderate and clustered along narrow zones in the S rift and E flank (Figure 2a). The largest (Md = 3.7, focal depth = 5 km) earthquake occurred on 22 September along the westernmost PF (Figure 2a). This was accompanied by surface transtensive fracturing on the PF, at 1450 m asl, showing sinistral offset ∼0.48 m (line 1, Figure 3a). Immediately thereafter, the mild strombolian summit activity, initiated three months earlier, ceased abruptly, with only sporadic ash emissions occurring through 26 October. At 20:25 GMT on 26 October, a 2 hours-long seismic swarm (>300 events with Md ≥1.0, 67 of them in dataset, with hypocenters <4 km bsl; Figure 2b) affected the summit area. Contemporaneously, a N-S striking, 1000 m long eruptive fissure opened on the S Rift, at 2850–2600 m asl (Figure 1); on early 27 October, right stepping en-echelon NE-SW eruptive fissures developed along the NE Rift, at 3010–1890 m asl. The opening of the NE Rift was accompanied by the downrift propagation of earthquakes. The obtained FPSs indicate strike-slip motions along NNW-SSE (dextral motion) or ENE-WSW (sinistral) planes consistent with an overall ∼ESE-WNW extension (Figure 2b). Few hours before the NE Rift eruption, the western PF started to reactivate (points 1 to 4, Figure 1), with transtensive sinistral offset ∼0.77 m (line 1; Figure 3a). At depth, these geometry and kinematics are consistent with the ∼E-W planes (sinistral motion) in the FPSs “B” (Figure 2b) and “D” (Figure 2c). On 29 October, a seismic swarm (hypocenters clustering at ∼4 km bsl) affected the Timpe Fault System (TFS) on the E flank (SV area in Figure 1), inducing NW-SE striking surface fractures with transtensive dextral displacement ∼0.04 m (TFS, Figure 1; line 7, Figure 3a). On this day, both the FPSs “E” and “F” indicate an ∼ESE-WNW extension along the E flank (Figure 2c). During early November the PF continued to deform: by 12 November surface fracturing, totalling ∼1.25 m of sinistral offset on the westernmost PF (line 1, Figure 3a), had propagated to the coastline (minimum offset 0.03 m, Figure 3a), through E-W striking fractures with predominant transtensive left-lateral shear (lines 5 and 6, Figure 3a). During November 2002–February 2003, seismicity was mainly located on the NE and, scattered, on the SE flanks (Figure 2d). On 26 November, minor seismicity at Trecastagni (TF in Figure 1) induced surface fracturing with a predominant transtensive dextral movement ∼0.035 m (line 8, Figure 3a).
 The opening directions of the transtensive sinistral and dextral faults of Figure 3 show, as indicated in Figure 1, an overall ESE-WNW trend. Volcanic activity ended in early November (NE Rift) and late January (S Rift).
4. Discussion and Conclusions
 Slow (<1 cm/yr) aseismic spreading had occurred at Etna, including creep-like movements along the PF, since ∼1994 [Froger et al., 2001; Bonforte and Puglisi, 2003; Lundgren et al., 2003]. Continued slow deformation is indicated by the moderate seismicity on the E and S flanks before September 2002 (Figure 2a). Within this framework, the September earthquake marks a major abrupt acceleration in the flank instability.
 Two lines of evidence suggest a correlation between the September crisis and the October eruption. (1) The cessation of the months-long summit activity, suggesting incipient lateral draining of magma from the central conduits. (2) The extremely short period (2 hours) of pre-eruptive seismicity related to magma ascent, contrasting with 4 days of premonitory seismicity before the 2001 eruption [Behncke and Neri, 2003]. This precludes a forceful dike uprise in 2002, suggesting that the eruption resulted from the passive rise of magma in response to flank sliding. This passive behaviour may account for the delay (34 days) between the earthquake and the onset of the eruption.
 In the two weeks after the eruption beginning the entire (≥18 km long) PF activated. The eastward decrease in displacement may be partly explained by the eastward increase of clayish deposits in the substratum, which induce diffuse and retarded creep [Tibaldi and Groppelli, 2002]. Surface deformation also extended to the TFS (29 October) and TF (26 November). The deformation thus affected two discrete blocks, bordered by the PF, TFS and TF, at different times and rates (Figure 3). Both the surface fractures (Figure 1) and FPS (Figure 2) during the 2002–2003 events indicate overall ESE-WNW extension; this is consistent with an ESE slip of the slide blocks, gradually decreasing to SE.
 These events suggest a connection between flank slip and volcanism, characterized by the following stages. (1) Surface rupturing on the western PF during the September earthquake (inception stage of slip, Figure 4a). (2) Increase of the deformation on the E flank and eruption along the NE and S Rifts (climax stage, Figure 4b). (3) Growth of the slipping area, with the activation of the TF, and slippage propagating to the adjacent slide block to the S, affecting a total area ∼370 km2 (propagation stage, Figure 4c). The distribution of earthquake hypocenters below the E flank (Figure 2c and d) suggests a mean décollement depth for the sliding blocks of 4 km bsl. This depth, different from what previously proposed [Bousquet and Lanzafame, 2001; Lo Giudice and Rasa, 1992; Borgia et al., 1992], corresponds roughly to the depth of sub-edifice magma accumulation [Patanè et al., 2003] invoked as a main cause of flank instability [Lundgren et al., 2003]. In this framework, the 2002–2003 events might represent the shallow effects of a much longer, deep-seated process.
 The authors thank R. Funiciello for encouragement, L. Ferrari, A. Tibaldi and an anonymous reviewer for useful comments.