The use of recombinant severe acute respiratory syndrome-coronavirus (SARS-CoV) nucleocapsid protein (N) enzyme-linked immunosorbent assay (ELISA)-based antibody and antigen tests for diagnosis of SARS-CoV infections have been widely reported. However, no recombinant SARS-CoV spike protein (S)-based ELISA is currently available. In this article, we describe the problems and solutions of setting up the recombinant SARS-CoV S-based ELISA for antibody detection. The SARS-CoV S-based immunoglobulin M (IgM) and IgG ELISAs were evaluated and compared with the corresponding N-based ELISA for serodiagnosis of SARS-CoV pneumonia, using sera from 148 healthy blood donors who donated blood 3 years ago as controls and 95 SARS-CoV pneumonia patients in Hong Kong. Results obtained by the recombinant S (rS)-based IgG ELISA using the regenerated S prepared by dialysis with decreasing concentrations of urea or direct addition of different coating buffers, followed by addition of different regeneration buffer, identified 4 M urea and 1 M sarcosine for plate coating and no regeneration buffer as the most optimal conditions for antibody detection. The specificities of the S-based ELISA for IgG and IgM detection were 98.6% and 93.9%, with corresponding sensitivities of 58.9% and 74.7%, respectively. The sensitivity of the rN IgG ELISA (94.7%) is significantly higher than that of the rS IgG ELISA (P < 0.001), whereas the sensitivity of the rS IgM ELISA is significantly higher than that of the rN IgM ELISA (55.2%) (P < 0.01). An ELISA for detection of IgM against S and N could be more sensitive than one that detects IgM against N alone for serodiagnosis of SARS-CoV pneumonia.