Special numbers overflow and underflow threaten the software security in programs or software. To find the overflow vulnerabilities features, we design a multi-dimensional structure according to IEEE 754 floating point special numbers formats. Based on special numbers definition in language and arithmetic operation overflow ranges, we analyze the overflow and underflow properties of special numbers arithmetic operation in algebra systems. We test the arithmetic overflow and underflow of floating point special numbers in different CPUs. The experimental results indicate the floating point overflow exists in many kinds of language programs and binary executable software.