Data mining using neural networks has been applied in various financial fields such as risk mitigation, missing data filling, fraud detection, and customer profile classification etc. This master thesis work aims to develop methodologies to mine large sets of records and in particular to fill missing data in these records. The steps include data cleansing, data selection, data preprocessing, data representation, data clustering and finally the missing data filling. Furthermore, this work designs algorithms to evaluate the supervised neural networks' performance, which is helpful for the future research on data prediction and classification. A case study based on a large data set of credit card records, which contains incomplete records, is performed to demonstrate that the proposed algorithms and their implementations accomplish the task of filling missing data in such records.