The onset of flooding is thought to be highly sensitive to the end geometry of the test channel and its high sensitivity is an important cause of data scattering. In the present work, the effect of end geometry on flooding is experimentally investigated. Observation shows that flooding definition also causes the data scattering and that flooding should be classified into two modes according to the location where it occurs. With the sharp liquid entrance geometry, both entrance flooding and exit flooding are able to take place while only exit flooding does when entrance is smooth. In the range of low liquid flow rates, exit flooding always takes place and flooding gas velocities for the same exit geometry are nearly the same regardless of liquid entrance geometry. When air is injected into the test tube through a nozzle, even if the liquid exit is sharp, flooding gas velocities are nearly the same as those with smooth exit geometry. The present experiments show that the length effect is significantly affected by the pipe-end geometry. It is small in the range of low liquid flow rates but becomes significant as the liquid how rate increases.