Data for 42 stations in different parts of the world in the northern hemisphere have been employed to partition monthly averaged daily global radiation (H) over bar and sunshine duration a in a bid to obtain improved fits to Angstrom＇s correlation. It has been found that regression fits to the correlation using data for biannual groups of months from March-August (months 3-8) and September-February (months 9-2), or March-September (months 3-9) and October-February (months 10-2), give an improvement in the rms error over the year, which is 25% or higher than the errors for annual fits for half of the cases. In no case is there an increase in rms error from the partitioning. It is found that biannual regression parameters for a pyranometer station may be used to predict with good accuracy global radiation for locations hundreds of kilometers away from the station if the climate, altitude and latitude are similar. A use of the seasonal partitioning of data leads to the following relations with station independent coefficients for (n) over bar/(N) over bar (H) over bar/(H) over bar(O), = 0.29 cos Phi + 0.49 (n) over bar/(N) over bar for months 10-2. and (H) over bar/(H) over bar(O), = 0.29 cos Phi + 0.54 (n) over bar/(N) over bar for months 3-9 These give better estimates for India than popular station independent formulae. It has been shown that if the coefficients of (n) over bar/(N) over bar are considered to be dependent on the climate as well, even more accurate estimates are obtained for the central and northern portions of the Indian subcontinent. The average rms is found to be 2.5%.