A detailed analysis of backbone amide NH chemical shift temperature gradients (Delta delta/Delta T values) for proteins and highly cross-linked peptides reveals that hydrogen-bonded exchange-protected NHs are characterized by Delta delta/Delta T values of -2.0 +/- 1.4 ppb/degrees C while exposed NHs typically display gradients of -6.0 --> -8.5 ppb/degrees C; however, numerous exceptions to these generalizations occur. For partially folded peptides (rather than proteins), exceptions are more common than concordance with this rule; Delta delta/Delta T values ranging from -28 to +12 ppb/degrees C have been observed. In the case of the peptide systems for which exchange protection data is available, the common practice of assuming that a Delta delta/Delta T value less negative than -4 ppb/degrees C indicates that the NH is sequestered from solvent is shown to have zero predictive validity. The analysis of the data for partially folded peptides, protein fragments, and other peptides which are expected to display minimal structuring reveals a significant correlation between Delta delta/Delta T and the deviation of delta(NH) from the random coil reference shift. The analysis was facilitated by plotting NH chemical shift deviations (NH-CSD) versus the Delta delta/Delta T values. Using such plots, slow-exchanging hydrogen-bonded sites in proteins can be determined with much higher confidence than using the value of the gradient alone. For peptides, the occurrence of large shift deviations and abnormal gradients are diagnostic for partial structuring at lower temperatures which becomes increasingly randomized on warming. A good correlation coefficient (R greater than or equal to 0.75) for NH-CSD and Delta delta/Delta T values indicates that essentially all of the NH shift deviation from reference values is due to the concerted formation of a single structured state on cooling. Correlation coefficients greater than 0.95 were observed for both helix and beta-hairpin forming peptides. The slope of the correlation plot (parts per thousand/degrees C) is a measure of the decrease in the population of the structured state upon warming. A detailed model which rationalizes the effects of conformational equilibria upon NH shifts is presented. A positive Delta Cp for unfolding is required to rationalize the linearity of delta(NH) With temperature that is routinely observed for partially structured peptides. This analysis suggests that ordered states of short peptides achieve significant populations in water only when the hydrophobic effect favors the structured state. This conclusion is pertinent to the current questions concerning the temporal sequence of secondary versus tertiary structure formation during protein folding. Further, it is suggested that the use of NMR parameters (scalar and dipolar couplings) to derive the structural preferences of protein fragments which might serve a ''seeding'' role in the folding pathway is justified only when the CSD/gradient plot displays both a correlation coefficient greater than 0.70 and significant NH-CSD values (\CSD\ > 0.3).