Several municipalities have recently installed wireless 'smart' water meters that allow functionalities such as demand response, leak alerts, identification of characteristic demand patterns, and detailed consumption analysis. To achieve these benefits, the meter data needs to be error-free, which is not necessarily available in practice, due to 'dirtiness' or 'uncertainty' of data, which is mostly unavoidable. The focus of this paper is to investigate practical solutions to mine uncertain data for reliable results and to evaluate the impact of dirty data on filters. This evaluation would eventually lead to valuable information, which can be used for educated decision making on water planning strategies. We perform a systematic study of the errors existing in a large-scale smart water meter deployments, which is helpful to better understand the nature of errors. Identifying customers contributing to a load peak is used as the main filter. The filter outputs are then combined with the domain expert knowledge to evaluate their accuracy and validity and also to look for potential errors. After discovering each error, we analyze its trails in the data and track back its source, which would eventually lead to the removal of the error or dealing with it accordingly. This procedure is applied progressively to ensure that all detectable errors are discovered and characterized in the data model. We evaluate the performance of the proposed approach using the smart water meter consumption data obtained from the City of Abbotsford, British Columbia, Canada. We present the results of both unprocessed and cleaned data and analyze, in detail, the sensitivity of the selected filter to the errors.