One of the main problems of observational cosmology is to determine the range in which a reliable measurement of galaxy correlations is possible .
This corresponds to determine the shape of the correlation function , its possible evolution with redshift and the size and amplitude of large scale structures .
Different selection effects , inevitably entering in any observation , introduce important constraints in the measurement of correlations .
In the context of galaxy redshift surveys selection effects can be caused by observational techniques and strategies and by implicit assumptions used in the data analysis .
Generally all these effects are taken into account by using pair-counting algorithms to measure two-point correlations .
We review these methods stressing that they are based on the a-priori assumption that galaxy distribution is spatially homogeneous inside a given sample .
We show that , when this assumption is not satisfied by the data , results of the correlation analysis are affected by finite size effects.In order to quantify these effects , we introduce a new method based on the computation of the gradient of galaxy counts along tiny cylinders .
We show , by using artificial homogeneous and inhomogeneous point distributions , that this method is to identify redshift dependent selection effects and to disentangle them from the presence of large scale density fluctuations .
We then apply this new method to several redshift catalogs and we find evidences that galaxy distribution , in those samples where selection effects are small enough , is characterized by power-law correlations with exponent \gamma = 0.9 up to 20 Mpc/h followed by a change of slope that , in the range [ 20,100 ] Mpc/h , corresponds to a power-law exponent \gamma = 0.25 .
Whether a crossover to spatial unformity occurs at \sim 100 Mpc/h can not be clarified by the present data .