Method 1 is the correct one. The baseline should be a line or a sigmoid ( i.e. two parallel lines with a transition zone between then, if the sample heat capacity is changing during the transition). I don't think there is a valid phenomena on which to base the second baseline.

The fact that the point between your two events does not reach the baseline is due to the superposition of the two thermal events. Nonetheless you can still calculate the Tonset for the second event by extrapolation.

Thank you very much Mitran

The method one is to be chosen. But be careful, the way you performed was not exactly correct since you draw lines between arbitrarily chosen points. The explanation provided by  Raul is very good, I would simply complete it by clarifying the way you obtain each parallel baseline you need to perform 2 fits, one before the transition and one after the transition. You will obtain 2 lines, supposedly parallel. So now yo need to define the points that are corresponding to what is "before the transition" and "what is after the transition". The easiest is to program a linear fit, with first point the start of your curve and last point the one that will provide the largest range without deviation in the Chi_square. Use a criteria as such as the residual should not exceed a predefined value. Then do same for 2nd zone. The accuracy of these fits will greatly help to reach best subtracted data. And then derived intersepts, enthalpies etc... will gain also in accuracy.

Ananda... The DSC you've shown indicates two thermal changes. The second one occurs before the end of the first. The best solution is to repeat the run at a smaller heating rate. You'll end up with a perfectly isolated peaks and the baseline will be self indicated. There will be no need for interpolation or approximation... Good luck 

You must choose the 1. There is no reason that justifies the use of the second option, where the results are altered and information is lost

Responses from Raul and Pierre are correct.

Mahmood's proposal would allow you to get a reliable result.

Yes, method 1, and yes, Mahmood Barbooti's suggestion should help.  However, if you find that a smaller heating rate still doesn't deconvolute the peaks sufficiently, you can try to repeat the run at several different smaller heating rates (e.g. 1/2, 1/4, 1/8 of the present one) and extrapolate the apparent onset of peak 2 to a heating rate of zero.  (You can't push this too far though, because the signal gets smaller and smaller with decreasing heating rates and with it the signal/noise ratio.)

Thank you very much Mitran, Braun, Panine, Barbooty, Marigliano, and Cooper. Your answers were outstanding and very useful.