miRNAs are small about 22-base pair long, RNA molecules are of significant biological importance. Like other longer RNA molecules, messages in miRNAs are words in an alphabet consists of only four nucleotide bases. However, just like words in any language, not all combinations of these alphabets are not meaningful. In fact, we find that the distributions of nucleotides bases in human miRNAs show significant deviation from randomness. First, a miRNA sequence containing four bases are mapped into a binary string with three kinds of classifications according to their chemical properties. Our analyses based on some statistical measurement clearly demonstrate that the purines-pyrimidines class of nucleotide bases resembles all the human miRNAs. Then, we propose a simple nearest neighbor model (Ising model) to understand the statistical variations in human miRNAs. One the way, we also discuss the limitations of a mean-field model.