Coal proximate analysis is a form of typical compositional data, commonly represented with constant sum. Although the direct geostatistical modeling of compositional data provides apparently reasonable outputs, the results are always exposed to inconsistency and non-optimality. In this paper, we compare the compositional and noncompositional approaches to assess the problems caused by neglecting the compositional nature of data. The ultimate goal is to attain an accurate approach for coal reserve evaluation. The presented compositional approach was executed and validated on field data from the Parvadeh IV coal deposit in Central Iran. The comparison of sum of values maps, Aitchison distance, actual vs. estimated cross-plots and correct classification rate of the mentioned approaches illustrate that the compositional approach is notably more accurate and reliable. To compare tonnage - grade curves, new formulations are proposed to calculate T and T+, two moments of total tonnage, defined based on the random variables of stochastic simulation. Tonnage - grade and mean grade - cutoff grade curves showed that the noncompositional approach has overestimated ash values and underestimated carbon content. This can lead the analyst to misinterpretation and underrating the deposit. Quantitative comparison of tonnage - grade curves of the approaches revealed that at cut-offs of 31% for ash and 52% for fixed carbon, nearly 5 and 3 million tons of coal are being considered as waste by the noncompositional approach. Consequently, neglecting the compositional nature of data will result in deviated outputs, unrealistic models and unreliable evaluations and finally lead to financial losses. Thus, it is strongly recommended to consider the compositional nature of data in reserve evaluations.