This paper presents a calibration framework based on the generalized likelihood uncertainty estimation (GLUE) that can be used to condition hydrological model parameter distributions in scarcely gauged river basins, where data is uncertain, intermittent or nonconcomitant. At the heart of this framework is the conditioning of the model parameters such as to reproduce key signatures of the observed data within some limits of acceptability. These signatures are either based on hard or on soft information. Hard information signatures are defined as signatures for which the limits of acceptability may be objectively derived from the distribution of long series of observed values, and which effectively constrain the model parameters. Soft signatures are less effective in parameter conditioning or their limits of acceptability cannot be objectively derived. During random parameter sampling, parameter sets are accepted as equally likely if they meet all the hard limits of acceptability. This results in an intermediate parameter distribution, which can be used to reduce the sampling limits. Then, the soft information may be introduced in a second constraining step to reach a final parameter distribution. The modeler can use the final results as a guideline for a further search for information, possibly from new observations yet to collect. In an application of the framework to the Luangwa catchment in Zambia, three information signatures are retrieved from a data set of old discharge time series and used to condition the parameters of a daily conceptual rainfall-runoff model. We performed two independent calibration experiments with two significantly different satellite rainfall estimates as model input. The results show consistent parameter distributions and considerable reduction of the prior parameter space and corresponding output realizations. These results illustrate the potential of the proposed calibration framework for predictions in scarcely gauged catchments.