Hypotheses explaining the use of intermittent bounding and undulating flight modes in birds are considered. Existing theoretical models of intermittent flight have assumed that the animal flies at a constant speed throughout. They predict that mean mechanical power in undulating (flap-gliding) flight is reduced compared to steady flight over a broad range of speeds, but is reduced in bounding flight only at very high flight speeds. Lift generated by the bird's body or tail has a small effect on power, but is insufficient to explain observations of bounding at intermediate flight speeds. Measurements on starlings Sturnus vulgaris in undulating flight in a wind tunnel show that flight speed varies by around ±1 m/sec during a flap-glide cycle. Dynamic energy is used to quantify flight performance, and reveals that the geometry of the flight path depends upon wingbeat kinematics, and that neither flapping nor gliding phases are at constant speed and angle to the horizontal. The bird gains both kinetic and potential energy during the flapping phases. A new theoretical model indicates that such speed variation can give significant savings in mechanical power in both bounding and undulating flight. Alternative hypotheses for intermittent flight include a gearing mechanism, based on duty factor, mediating muscle power or force output against aerodynamic requirements. This could explain the use of bounding flight in hovering and climbing in small passerines. Both bounding and undulating confer other adaptive benefits; undulating may be primitive in birds, but bounding may have evolved in response to flight performance optimization, or to factors such as unpredictability in response to predation.