Hi,Dear Salehi

please have a look at this site.

http://www.ni.com/white-paper/7112/en/

Read what Jerónimo Quesada said above and then generate a callibration curve with 3 or 4 points (and several measuremts at each point).

You migt also try with paralel  resistor switched to NTC. You will obtain less temperature senstitivity  but simultaneousoly better linearity of the temperature characteristics. I think , best if you apply standard  PT100 sensor instead of NTC together with  appropriate instrumentation calibrated amplifier usually available  from any  supplier of automation equipment.    

This maker of NTC controllers has some good info on the subject in this document:

http://sites.fas.harvard.edu/~phys191r/Bench_Notes/A10/ldt5412.pdf

It is usually best to accurately record resistance, being careful  to avoid heating the thermistor with the measurement current, and then linearize and calibrate digitally.

Also consider if NTC is the best choice, many electronic sensors are quite linear of the shelf. Thermocouples and  Pt100 are better alternatives for many applications. Thermistors are often a good choice for sensitive differential measurements, but less so for high absolute accuracy.

If the  0.5C temperature measurement accuracy is acceptable, I recommend   TMP275 or TMP112 digital output temperature sensor  from Texas Instruments. I think you will get best cost-to-performance index.

Hi,Dear Salehi

please have a look at this site.

http://www.ni.com/white-paper/7112/en/