Using both household level and plot-level data from Northern Bangladesh, this paper analyzes the difference in agricultural productivity across different contractual arrangements among ultra-poor households. Employing the fixed-effect model on the pseudo panel data, the paper finds evidence of sub-optimal use of inputs and, consequently, lower productivity for lands cultivated under sharecropping contracts. The inefficiency on the part of sharecroppers is also evident from the stochastic frontier model. Although the paper finds that credit has no direct impact on productivity, the results of Logit estimates suggest that availability of credit induces households to opt for fixed-rental contracts.