the systematic design stands out for its compactness, broadness and for allowing the testing of a larger number of spacings. however, it is not used due to the systematic arrangement (non-randomized) of the plants and the high sensibility for missing values. the aim of this work was to describe the geostatistic model and associated methods of inference in the analysis context of non-randomized experiment, reporting applied results to identify the spatial dependence in a fan systematic design of eucalyptus dunnii. furthermore, different alternatives for treating missing values that can occur from flaws and/or mortality of plants were proposed, analyzed and compared. data were analyzed by three models that differed, with covariates, in the form of modeling missing data values. a semivariogram was built for each model, adjusting three correlation function models, being the parameters estimated through the maximum likelihood method and selected by the akaike's criterion. these models, with and without the spatial component, were compared by the likelihood ratio test. the results showed that: (1) the covariates interacted positively with the response variable, avoiding data to be discarded; (2) the model comparison, with and without the spatial component, did not confirm the existence of dependence; (3) the incorporation of the spatial dependence structure into the observational models recovered the capacity to make valid inferences in the absence of randomization, overcoming operational problems and guaranteeing that the data can be subjected to classic analysis.