This paper attempts to assess the link between poverty and efficiency in Ethiopia. Based on 214 households drawn from eight Peasant Associations (PAs) from two zones in Ethiopia’s Amhara National Regional State, the study measures the levels of poverty and efficiency, and assessed the link between the two. Both objective and subjective measures were used to measure the level of poverty, which was found to be 28% and 27% by both measures, respectively. While 35.6% of households were found to be poor in relative terms (using the Cost-of-Basic-Needs approach), 76.4% were poor in terms of the one-dollar-a-day absolute measure of poverty. The study also shows that there was wide variation in the level of poverty between the sample districts. The stochastic frontier distance function with variable return to scale specification reveals that all input variables, with the exception of ox power, significantly determine the efficient level of output. Whereas the average efficiency score of sample households was 62.8%, indicating the existence of technical inefficiency among farmers in the study areas, the mean efficiency score of poor and non-poor households was found to be 58.3% and 74.4%, respectively. The logistic regression also reveals that efficiency plays an important role in explaining the observed poverty incidences. Prediction of the marginal effects of a 10% improvement in efficiency reduced the probability of being poor by about 4%, while simulation of the effect of improved efficiency on poverty indicates that 14.1% of the poor households could get out of poverty by attaining a 25% improvement in the existing efficiency levels. The implication for policy is that improving access to resources without ensuring their efficient utilization could not have the desired impact in reducing poverty.