The paper presents results of unemployment estimation for small domains in Poland. The research was undertaken within the EURAREA project (IST-2000-26290), its main objective being to assess the effectiveness of 'standard' small area estimation techniques in the statistical systems of six European countries. Properties of indirect unemployment estimation at local labour markets in Poland have been studied for the proportion of ILO unemployed at two different levels of territorial division NUTS3 and NUTS4. The main objectives of the research were: to test how the standard Small Area Estimators would perform on Polish database and to determine their area specific properties, next to make evaluation for all areas, and to analyse of characteristics of the estimators' distribution and general estimators characteristics. The EURAREA approach of simulation using the 'standard' set of estimators is presented and applied.The properties of the estimators are discussed from the domain specific point of view and combining all areas. Conclusions drawn from simulation results can be synthesised as follows. The results confirm the basic properties of direct estimators: they are unbiased and have a high variance. Just oppositely, the indirect estimators have small variance but might be seriously biased when the assumptions are not met. The composite estimators serve as a compromise between the ineffectiveness of the direct estimation and a bias of synthetic approach.High values of relative estimation errors, which at the NUTS4 level ranged between 25-40%, seem to be unsatisfactory. Results obtained for NUTS3 are much better as the average value of the Relative Estimation Error does not exceed 25%. Nevertheless the measures of efficiency for small area estimation show an impressive gain in estimation precision when comparing with direct estimates. For NUTS4 the best results found for the synthetic and EBLUP estimators with area level covariates show the Mean Square Error about 85% smaller than the variance of the direct estimator. The gain obtained for NUTS3 level is also remarkable: it amounts to about 50% when applying the EBLUP estimator with area level covariates instead of direct one. However, estimation results are not stable for the area level as the analysis of the Relative Estimation Error distributions demonstrates. The biggest stability was observed for the EBLUP estimators. Their properties: rightward skewness, one peak and big kurtosis are also more preferable. The results of the study present the practical possibilities and benefits of adopting the techniques of small area estimation of unemployment in Poland.