try this paper from this link

http://www.researchgate.net/publication/228791287_Comparison_of_Regression_Kriging_and_Cokriging_Techniques_to_Estimate_Soil_Salinity_Using_Landsat_Images/file/9c96051eaf38176f4d.pdf.

Article Comparison of Regression Kriging and Cokriging Techniques to...

ARCGis is not a geostatistical tool or method however there is an add-on for ARCGis for spatial analysis that includes various geostatistical tools, see http://www.esri.com/library/whitepapers/pdfs/geostat.pdf for a description of the methods included in "Geostatistical analyst". I assume that by "areal interpolation" you are referring to estimation of average values over subregions, typically rectangular (that restriction is a programming quirk rather than a theoretical problem). In the geostatistical literature this is known as "block kriging". This is also implemented in various other geostatistical software programs, e.g. see  the R package "Geostat".

Co-kriging pertains to multivariate data sets, i.e. where you have data for several different attributes at the same locations but not necessarily for all attributes at all data locations. Often it is used when you have less data for a primary attribute but more data for one or more correlated secondary variables. Co-located cokriging is a variation on this form of cokriging. See books by H. Wackernagel and by M. Hohn

Interesting comment from @Donald. Here usually use the R package "geoR".

Thanks sir for your reply. I have already used this to create a prediction surface. But,If i have not bolck data, how can i make this? Is there any reference to create block data map over the polygons?