Nitrogen (N) and phosphorus (P), especially the N in Rubisco that drives photosynthesis and the P in rRNA that drives the generation and maintenance of proteins, are essential nutrients for plants. As an important plant leaf trait and allometric “rule” in ecology, the scaling relationship between leaf N and P concentrations is crucial to modelling N and P cycles in terrestrial ecosystems. Previous studies have generalized an invariably “constant” law that N scales roughly as the 2/3 or 3/4 power of P (i.e., N∝Pα=2/3 or 3/4). However, whether the numerical value of the scaling exponent is constant remains unclear and is one of key issues in plant ecology. To address how the numerical value of the scaling exponent changes with functional groups and environmental conditions, we compiled a global data set and found that the exponent varied significantly across different functional groups, latitudinal zones, ecoregions (continents), and sites. The exponents of herbaceous and woody plants were 0.659 and 0.705, respectively. Among woody plants, the exponents of coniferous, deciduous and evergreen broad-leaved species were 0.610, 0.712 and 0.731, respectively. The exponents also showed significant latitudinal patterns, decreasing from tropical to temperate to boreal zones. Further, across the ecoregions of North America, Europe, Asia, Oceania, Africa, and South America, the exponents were 0.603, 0.672, 0.712, 0.786, 0.835, and 1.071, respectively. At sites with a sample size >10, the values fluctuated from 0.366 to 1.928, with an average of 0.841. Such large numerical variations of the N vs. P scaling exponents likely reflect species composition, P-related growth rates, relative nutrient availability of soils and a number of other factors. Our results therefore indicated that there is no canonical numerical value for the leaf N vs. P scaling exponent. The traditional analysis of pooled data at global scale for this scaling relationship hides biologically and ecologically significant variation. This finding has a critical bearing on the parameterization of N and P biogeochemical models.