This paper describes the behavioral data of an experiment in which water flows (WFs) were first used as replacement of the traditional electric shocks to test free-operant avoidance in Betta splendens (Hurtado-Parrado et al. 2019 https://doi.org/10.1016/j.beproc.2018.10.021). WFs with a duration of 10 s each were delivered with 30-s flow-flow (F-F) and response-flow (R-F) intervals in a custom-made shuttle tank. Fish escaped or avoided the WFs by changing compartments. Crossings during the WFs, interrupted the flows, were automatically scored as escape (Esc), and initiated a new R-F interval. Crossings that occurred during R-F or F-F intervals were scored as avoidance responses and also reset the R-F interval. We compared the effect of adding a warning stimulus - curtains of air bubbles - to the last 5 s of the R-F interval; i.e., signaled versus unsignaled avoidance. A unique development of the WFs procedure, and thus the data here described, is that crossings were further differentiated into subcategories; namely, early avoidance (EA) if a crossing occurred during the first 25 s of the R-F interval; late avoidance (LA) if a crossing occurred during the last 5 s of the R-F interval; and Flow-Flow avoidance (FF) if a crossing occurred anytime during the F-F interval. Here we present the data of six bettas across the different phases of the experiment; namely, baseline (BL - no WFs programmed), signaled avoidance (SA - warning stimulus scheduled), and unsignaled avoidance (UA - no warning stimulus scheduled). The dataset available at the Open Science Framework (OSF) repository (http://doi.org/10.17605/OSF.IO/FMHXD (Hurtado-Parrado et al., 2019)) includes for each fish and per 20-min daily session the total number of crossings; frequency of each type of crossing (Esc, EA, LA, FF); total WF frequency and duration, the total time spent in each compartment, and an index of preference for each compartment based on the proportion of time spent in the tank's compartments.