Abstract: The fishway collecting system has been confined to complex flow patterns, diverse operating conditions, and unclear behavioral responses of Chinese carp fish. In this study, an experimental apparatus was constructed to emulate the main structure of a typical collecting system, according to the prevalent size and operational conditions of several fishways in China. 11 sets of tests were carried out under four working conditions. Two water levels and two outflow patterns were collected to observe the attractive rate, attempted rate, and upstream passage rate of mature crucian carp (Carassius auratus) at different spawning stages. Passive integrated transponder (PIT) telemetry within a generalized model was employed to monitor the fish swimming trajectories in the collecting system. Initially, the flow patterns were measured under the four working conditions. The flow velocity at low levels exceeded the high-water levels. The dispersed outflow exhibited greater continuity than the concentrated outflow through water holes from the supplemental water channel to the collecting canal. The experimental results indicated the attractive rates ranged from 5.24% to 94.44%, 41.46% to 92.59%, and 0 to 77.33%, respectively, at the three entrances. The random forest model demonstrated that five predictor variables (spawning stage, outflow condition, level, flow discharge, momentum, and velocity) were used to explain 60.46% of the density variation in the rates of attraction of crucian carp at the entrances, with the better match between predicted and observed values. Spawning stage and water momentum were ranked the first and the second among all environmental variables. Nonparametric statistical tests indicated that there was a significantly higher attractive rate before spawning than after spawning (P < 0.05). There was no significant difference between the small and medium momentum groups (P > 0.05). But both showed a more significant difference, compared with the large momentum group (P < 0.05). The binary logistic regression model revealed that the attempted rate before spawning was 33.032 times that after spawning, while the dispersed outflow from the water replenishing holes was 4.228 times that of centralized outflow, and the higher water level was 2.484 times that of the low water level. Bivariate correlation analysis showed that the upstream passage rate increased with the attempt rate (P < 0.05). The upstream behavior of fish in the powerhouse collecting systems was primarily driven by ecological needs and also influenced by the outflow flow. Momentum was used to comprehensively reflect both flow velocity and flow rate, thereby serving as the better indicator of the rate of attraction, in terms of water flow. However, the higher momentum cannot always lead to higher rates of attraction and is influenced by the ecological habits of fish and their life history stages. At the connection section from each entrance to the main channel of the fishway, continuous guided water flow greater than the attractive speed should be provided to guide the fish to successfully enter the main structure of the fishway and avoid turning back downstream. These indicators should be comprehensively considered in future designs to predict the upstream strategy of fish.