An investigation of the statistical properties of the native conformations of proteins, observed from crystal structures, is reported. Protein conformations were analyzed in terms of a bond vector correlation function and molecular volume. It was observed that, while the volume of a protein structure varies nearly linearly with the number of residues, the bond vector correlation function exhibits a universal feature for all sizes of proteins. To interpret the nature of the bond vector correlation function of native protein structures quantitatively, Monte Carlo simulations of realistic polypeptide chains of specific but arbitrary amino acid sequence were carried out. The molecule was constrained in an ellipsoidal volume determined by its chain length, and conformations with unacceptable nonbonded contacts between different amino acid residues were excluded. The interactions within a terminally blocked single residue, which correlate two nearest-neighbor peptide groups in a chain, were taken into account by an energetically biased sampling of its phi-psi space. The simulated chain correlation functions were found to be in good agreement with those of the crystal structures of beta-sheet-type and mixed-type (alpha+beta) proteins of similar length. On the basis of these calculations, it is concluded that the observed conformations of these native proteins may arise from two basic factors: the compactness of structures under hydrophobic interactions and the intrinsic stiffness of polypeptide chains due to the interactions within each terminally blocked residue.