Abstract: This paper discusses the unreliability of extreme value statistical inference,its presentation,causes and ways of improving the reliability,practical methods and reliability evaluation.It advances the possibility of applying the phenomenon of super-fitting,sampling error,the domain of convergence criteria,the maximum likelihood estimation of refined estimates and rough estimates,the pan-likelihood estimation,multi-parent phenomenon,Monte Carlo stochastic simulation concepts or methods; and obtained reliable results of the extreme value statistical inference for the multiple meteorological elements of various weather station.The theory of small-sample statistics is initially formed.