Hashemi Gazar, A., Javaherian, A. and Sabeti, H., 2011. Analysis of effective parameters for semblance-based coherency attributes to detect micro-faults and fractures. Journal of Seismic Exploration, 20: 23-44. Coherency attributes are useful in the interpretation of seismic data and can be applied to 3D seismic data. When coherency attributes are applied to seismic data, they indicate the continuity between two or more traces within a seismic window. The rate of seismic continuity is an index of geological continuity. Areas of traces that change with faults or other geological phenomena have lower coherency in comparison with adjacent traces. Coherency attributes can be divided into three major groups: (1) cross-correlation, (2) eigenstructure and (3) semblance. In this paper, first, the ability of the three coherency attributes mentioned above to detect micro-faults was tested over 3D real data. The results proved that the semblance algorithm was much more powerful than the other algorithms in detecting micro-faults. Therefore, in the remainder of the study only the semblance attribute was employed. The effect of the dominant frequency, the signal-to-noise ratio, the dimensions of the analysis cube, and the apparent dip in the x- and in the y-directions on the semblance coherency attribute was investigated. The effects of these parameters were tested on 3D synthetic seismic data consisting of (1) horizontal layers, (2) dipping layers, and (3) cross-dipping layers. It is shown that for frequencies up to 20 Hz, there was no clear image of the micro-faults. However, for frequencies above 20 Hz, the resolution of micro-faults was increased. The results indicate that micro-faults are detectable with a signal-to-noise ratio of 1 or higher. When a signal-to-noise ratio of 0.5 is selected, micro-faults can still be detectable but with 0963-0651/11/$5.00 © 2011 Geophysical Press Ltd. 24 HASHEMI GAZAR, JAVAHERIAN & SABETI a lower resolution. According to synthetic models, a temporal window of 32 ms (k = 8) showed the best results for horizontal and dipping layers. The best size for the spatial window is 10x10 for horizontal and dipping layers. Therefore, the optimum cube dimensions of analysis are 10x10 x8. For these dimensions, the signal-to-noise ratio increases and micro-faults are clearly detectable. Regarding the cross-dipping model, apparent dip directions, p and q, were analyzed. The same optimum value of 10 ms/m was obtained for both. Real data which is related to carbonate units showed satisfactory results as well: micro-faults and minor fractures hidden in the primary data were detectable after applying the algorithm.