Terminal velocities of liquid drops (of varying diameters) in air have been estimated by the method of neural nets using molar refraction (RM) and surface tension at 293 K (sigma (293)) as inputs. Application of the method to the Velocity versus drop diameter data on ten liquids including water at 103 points, of which nine liquids were used for training and one liquid for testing, yielded an average absolute deviation of 7.4% for the trained data and 7.7% for the tested data. Comparison with the conventional method based on 5 different correlations for different ranges of Eotvos number or Morton number and for different liquids has shown that the absolute average deviation is 39.3% for nine liquids and 41.7% for the tenth liquid, being much inferior to the performance of the neural network based predictions. Testing of the method further with the data reported by several investigators on water has shown an absolute average deviation of 10.5%, which is comparable to that of the conventional method (8.2%). Besides, the range of errors encountered in the proposed method has been much narrower compared to that of the conventional method. The present study showed that the proposed method of using only two easily estimated/commonly reported inputs (R-M and sigma (293)) can be considered as a sound alternative to the commonly reported approach based on dimensionless numbers (like Reynolds number, Eotvos number and Morton number), which needs information on a number of temperature-dependent properties (like viscosity, density and surface tension) both in terms of better performance as well as the ease of use.