Liver disease may become ameliorated in some patients with chronic hepatitis D virus (HDV) infection. We present here a study based on longitudinal sampling to investigate the viral dynamics in chronic HDV infection. We examined the HDV variants from different time points, especially those before and after the elevation of serum aminotransferase levels. The datasets from each patient were tested for positive selection by using maximum-likelihood methods with heterogeneous selective pressures along the nucleotide sequence. An average of 4.9%, ranging from 3.1 to 6.8%, of the entire delta antigen sites was regulated by a diversifying selection. Most of the positively selected sites were associated with immunogenic domains. Likelihood ratio tests revealed a significant fitness of positive selection over neutrality of the hepatitis delta antigen gene in all patients. We further adapted a neural network method to predict potential cytotoxic T ligand epitopes. Among the HLA-A*0201 cytotoxic T ligand epitopes, three consistent epitopes across all three genotypes were identified: amino acids (aa) 43 to 51, 50 to 58, and 114 to 122. Three patients (60%) had sites evolving under positive selection in the epitope from aa 43 to 51, and four patients (80%) had sites evolving under positive selection in the epitope from aa 114 to 122. The discovery of immunogenic epitopes, especially cytotoxic-T-lymphocyte ligands, associated with chronic HDV infection may be crucial for further development of novel treatments or designs in vaccine for HDV superinfection.