This paper aims to provide a new step towards statistical inference based on symbolic dynamics. Relevant null hypotheses can be expressed in probabilities associated with a given set of symbols, and therefore new nonparametric tests can be developed. Symbolic analysis of the dynamics of the second moment of a stochastic process has lead us to present a new semi-parametric test of heteroskedasticity. The test is fairly general as minimal assumptions are required to be used. Monte Carlo experiments mainly show the great power of the test and the fact that is very competitive test as compared with other powerful tests currently available in the relevant literature.