Benchmark calculations for several HTTR core states were performed with four cross-section sets which were generated from JENDL-3.3, JENDL-3.2, ENDF/B-VI.8 and JEFF-3.0 using a continuous energy Monte Carlo code MVP. The core states were a critical approach in which an annular core was formed at room temperature and solid cores at room temperature and at full power operation. Study of keff discrepancies caused by difference of the nuclear data libraries and identification of nuclides which have large effects on the keff discrepancies were carried out. Comparison of the respective keff from calculations and experiments was also carried out. As the results, for each of the HTTR core states, JENDL-3.3 yields a keff agreeing with the experiments within 1.5%Δk, JENDL-3.2 yields keff agreement within 1.7%Δk, and ENDF/B-VI.8 and JEFF-3.0 yield keff agreement within 1.8%Δk. There is little keff discrepancy between ENDF/B-VI.8 and JEFF-3.0. The keff between JENDL-3.3 and JENDL-3.2 is caused by difference of 235U data and has temperature dependency. The keff discrepancy between JENDL-3.3 and ENDF/B-VI.8 or JEFF-3.0 is mainly caused by difference in graphite data.