The aim of the present study was to identify a model for the blood-brain barrier based on the use of a continuous cell line, and to investigate the specificity of this model. A set of test compounds, reflecting different transport mechanisms and different degrees of permeability, as well as different physiochemical properties was selected. In vivo data for transport across the blood-brain barrier of this set of test compounds was generated as part of the study using two different in vivo models. A computational prediction model was also developed, based on 74 proprietary Pharmacia compounds, previously tested in one of the in vivo models. Molsurf descriptors were calculated and 21 descriptors were correlated with log(Brain(conc.)/Plasma(conc.)) using partial least squares projection to latent structures (PLS). However, the correlation between predicted and measured values was found to be rather low and differed between one and two log units for several of the compounds. The test compounds were analyzed in vitro using primary bovine and human brain endothelial cells co-cultured with astrocytes, and also using two different immortalized brain endothelial cell lines, one originating from rat and one from mouse. Cell models using cells not derived from the blood-brain barrier, ECV/C6, MDCK and Caco-2 cell lines, were also used. No linear correlation between in vivo and in vitro permeability was found for any of the in vitro models when all compounds were included in the analysis. The highest r2 values were seen in the bovine brain endothelial cells (r2=0.43) and MDCKwt (r2=0.46) cell models. Higher correlations were seen when only passively transported compounds were included in the analysis, bovine brain endothelial cells (r2=0.74), MDCKwt (r2=0.65) and Caco-2 (r2=0.86). By plotting in vivo Papp values against logDpH7.4 it was possible to classify compounds into four different classes: (1) compounds crossing the blood-brain barrier by passive diffusion, (2) compounds crossing the blood-brain barrier by blood-flow limited passive diffusion, (3) compounds crossing the blood-brain barrier by carrier mediated influx, and (4) compounds being actively excreted from the brain by active efflux. Papp and Pe values obtained using the different in vitro models were also plotted against logDpH7.4 and compared to the plot obtained when in vivo Papp values were used. Several of the in vitro models could distinguish between passively distributed compounds and efflux substrates. Of the cell lines included in the present study, the MDCKmdr-1 cell line gave the best separation of passively and effluxed compounds. Ratios between AUC in brain and AUC in blood were also calculated for six of the compounds and compared to ratios between Pe or Papp for transport in the apical to basolateral and basolateral to apical direction. Again the MDCKmdr-1 cell line gave the best correlation with only one compound (AZT) giving large discrepancy between in vitro and in vivo data. None of the in vitro models could identify compounds known to be substrates for carrier mediated influxed as such, and the results indicate that a tighter in vitro blood-brain barrier model probably is needed in order to facilitate studies on carrier mediated influx. The findings presented also indicate that identification of "batteries" of in vitro tests are likely to be necessary in order to improve in vitro-in vivo correlations and to make it possible to perform acceptable predictions of in vivo brain distributions from in vitro data.