Sensor networks are typically unattended because of their deployment in hazardous, hostile or remote environments. This makes the problem of conserving energy at individual sensor nodes challenging. S-MAC and PAMAS are two MAC protocols which periodically put nodes (selected at random) to sleep in order to achieve energy savings. Unlike these protocols, we propose an approach in which node duty cycles (i.e sleep and wake schedules) are based on their criticality. A distributed algorithm is used to find sets of winners and losers, who are then assigned appropriate slots in our TDMA based MAC protocol. We introduce the concept of of energy-criticality of a sensor node as a function of energies and traffic rates. Our protocol makes more critical nodes sleep longer, thereby balancing the energy consumption. Simulation results show that the performance of the protocol with increase in traffic load is better than existing protocols, thereby illustrating the energy balancing nature of the approach.