Block adjustment in aerial triangulation has been increasing its importance in practical use. Discussions on its accuracy and efficiency, however, have been mostly done from the view point of the technical treatment in the electronic computers, or iteration procedure of the numerical approaximation.Several discussions outside of the above view point are tried on this paper, treating them as the problem of geometrical transformation and theory of errors. Physical problems are not discussed here and mechanical errors are duely not considered.The principle of aerial triangulation is, from the geometrical point of view nothing but the pure Euclidia n transformation from the model coordinates to the ground geodetical coordinates, except the projective transformation which is applied at the first stage of the procedure, i. e. during the relative orientation. Other transformation formula such as higher order transformations are frequently applied for the purpose of reducing the discrepancies at the control points and tie points, but they have rather weak theoretical base compared to the so called Helmert transformation in Euclidian geometrical space.As for the accuracy obtained through the bloc adjustment of aerotriangulation, the errors are originating from the observation erros of the earlier steps of the procedure, such as relative orientation or point observations, and are propagated according the law of propagation of erros throughout the procedure. The resultant erros are adjusted and their standard deviations are computed as the criterion index of the obtained accuracy.These computation should be carried out applying the reduced values of the freedom of the system which are given as the number of observations subtracted the number of applied conditions.The accuracy of block adjustment should be evaluated from the residual erros at ground control points, not from the discrepancies at tie points only, because the final purpose of the block adjustment is to obtain the ground geodetic coordinates of the observed points, and not the strip coordinates nor block coordinates. Discrepancies at tie points can be used only as the criterion data for the individual strip adjustment or conditions to be used during the adjustment.Few schematical examples are considered to get the general idea of the above mentioned character of the block adjustment. The limitting size of a block may be less than 10 strips with about 10 models in each strip under the normal condition such as the Specification for the National Large Scale Mapping Project of Japan, which can be derived from the numerical considerations on the freedom of the whole block. And the reduction of errors as the results of applying the tie points conditions into the error equation are estimated no more than 20% considering the freedom of the block as well.