Aimed at the data missing problem in data quality, a relational Markov model (RMM) based approach was proposed, which combined RMM and the dynamic attribute selection (DAS) method to estimate missing values, taking into  full account the relations between attributes and making maximum use of available information in complete cases to improve the estimation performance of missing values. This approach utilized the relational Markov model to compute the transition probabilities from source to target state, and fills in missing values using the maximum posterior probability (MaxPost) and probability proportional (ProProp) methods. Comparative experiments on well-known datasets verify the effectiveness and advantage of this approach.