This paper considers the nonparametric estimation of spectral densities for second order stationary random fields on a d-dimensional lattice. I discuss some drawbacks of standard methods, and propose modified estimator classes with improved bias convergence rate, emphasizing the use of kernel methods and the choice of an optimal smoothing number. I prove uniform consistency and study the uniform asymptotic distribution when the optimal smoothing number is estimated from the sampled data.Keywords: Spatial data, spectral density, smoothing number, uniform asymptotic distribution, Bootstrap. AMS 2000 subject classifications:. Primary 62M30; secondary 62M15, 62G20 * A version of this paper is published in TEST, by Springer Verlag, DOI 10.1007/s11749-007-0059-5. I wish to thank Professor C. Velasco and two anonymous referees for their helpful comments and suggestions, and Professor P. M. Robinson for introducing me in the topic.