In Case-Based Reasoning (CBR), case base maintenance algorithms remove noisy or redundant cases from case bases. The best maintenance algorithm to use on a particular case base at a particular stage in a CBR system's lifetime will vary. In this paper, we propose a meta-case-based classifier for selecting the best maintenance algorithm. The classifier takes in a description of a case base that is to undergo maintenance, and uses meta-cases-descriptions of case bases that have undergone maintenance-to predict the best maintenance algorithm. For describing case bases, we use measures of dataset complexity. We present the results of experiments that show the classifier can come close to selecting the best possible maintenance algorithms.