In this paper, the theory of nonlinear dynamical system was used to investigate the dynamical behaviour and predictability of monthly mean sunspot relative number in modern epochs (January 1850 – May 1992). A fractal dimension D = 2.8 ± 0.1 was obtained. The results show that the monthly mean number is a complex chaotic system of low dimension, describable by a finite number (between 3 and 7) of parameter variables. Average time scale of predictability of the monthly mean was found to be 150 months. It is proposed that the scale-free region is best determined on the basis of residuals in least-squares linear fit. The effect of data size and noise on the computed fractal dimension is briefly discussed.