In our previous paper [5], we have obtained a decomposition of \f\, where f
is a function defined on R-d, that is analogous to the one proved by H. Ta
naka in the early sixties for Brownian martingales (the so-called 'Tanaka f
ormula'). The original proofs use purely analytic methods (e.g. the Caldero
n-Zygmund theory, etc.). In this paper, we give a new proof of our 'Tanaka
formula in analysis', that is based on probabilistic arguments. The main to
ols here are Brownian motion, stochastic calculus and Burkholder-Gundy ineq
ualities for martingales. These methods allow us to improve somewhat our pr
evious results, by proving that some significant constants do not depend on
the dimension d.