In most local earthquake tomography (LET) studies the data are selected according to the azimuthal-gap (GAP) criterion, which means rejecting all the sources outside the station network perimeter. In this article I show that in some cases this criterion is inappropriate and can be a reason for significant decimation of the relevant data, in turn leading to lower quality inversion results. This study presents several qualitative and quantitative arguments for why the GAP criterion is not adequate. The fact of great importance to out-of-network events for improving the results of tomographic inversion is supported by synthetic testing using realistic distribution of events in the area of Central Java and station locations according to the Merapi Amphibious Experiments project. I consider three models with different criteria of event selection: (1) dataset with GAP<180°, (2) dataset with GAP<280°, and (3) dataset with all events within a radius of 5°. The synthetic modeling reproduces the real situation when neither coordinates of sources nor starting 1D models are initially known. The reconstruction results show that the best resolution is obtained for model 3 with all data available, while the worst solution is observed in model 1. This study demonstrates that the commonly used GAP criterion that rejects the out-of-network events is injurious for LET tomography. In future experiments and when reconsidering the old datasets, I encourage the use of data from all events to improve the results of tomographic inversion, though at large distances from networks (at least, up to 400–500 km).