This work deals with econometric modeling of bilateral trade flows based on gravity model. Standard approach in most of previous empirical researches consisted of estimating bilateral trade potentials using gravity model and analysis of differences between the observed and predicted (potential) trade flows. Large differences were interpreted as the unexhausted foreign trade potentials. This work considers some limitations and problems of such approach mostly based on cross-section data. We consider alternative gravity model specifications with panel data and estimating procedures, as appropriate base for more precise estimates and conclusions. Furthermore, both theoretical and empirical analysis of econometric problems in panel data gravity model are carried out. Some of those problems have considered partially in previous empirical researches (for example autocorrelation in panels), but some of them have not considered at all, such as double endogenous regressors. Empirical results show that mentioned problems cause biased regression parameters estimates and consequently systematic variations of gravity model residuals (large systematic differences between observed and predicted). This makes conclusions on trade potentials between countries imprecise and unreliable.