Flood basalt emplacement results from eruption of dyke swarms. Several lines of evidence suggest, however, that a large number of dykes in large igneous provinces have never erupted, or have done so over too short a length for any evidence of eruption to be found in the field. We have mapped 1025 dykes and dyke segments from two swarms in western Ethiopia, most of which are of basaltic composition, the others of silicic composition. The dykes cut across the basal breccia of the trap series, and are observed at a structural level c. 1000 m below the current top of the lava pile on the nearby volcanic plateau. We test the hypothesis that at such a depth, some of the basaltic dykes may have erupted whereas some others may not. Due to higher buoyancy within the trap series, silicic dykes are likely to have been systematically eruptive. The area is flat, which has allowed analysis of dyke length distribution at constant structural level, a useful approach for the characterisation of other fracture types such as tension fractures and faults. We found that dyke length follows a power-law distribution. When en échelon dyke segments are counted as single dykes (having as their length the accumulated length of the individual segments), the population exponent, 1.8, is in the mean of the exponents usually obtained for fracture data sets. This exponent correctly fits the shortest dykes considered (1–9 km). A much steeper slope (3.3) is obtained for the distribution of intermediate lengths (9–10 to 10–20 km), and is interpreted as a possible consequence of the eruption of those dominantly basaltic dykes that are more than 9–10 km long at the observed structural depth. The longest, silicic dykes (10–20 to 50 km) have the same population exponent as the shortest, basaltic dykes. This is explained by the homogeneous propagation mode of both populations, either dominantly horizontal or dominantly vertical. However, the silicic dyke population exhibits a slightly larger population coefficient, for which we discuss the possible influence of the Precambrian fabric. Analysis of fracture length distribution appears to be a promising tool for estimating how much of a dyke swarm is eruptive and better understanding the mechanisms of fissure eruption in large igneous provinces.