That really depends on what kind of relationship you think there should be between the two quantities being plotted.
Since fitting to a straight line is a "classic" problem in regression analysis, it is conventional to transform the quantities into something where a linear relationship is expected. For example, when plotting reaction rate versus temperature, where an exponential-type relationship is expected, you usually plot the logarithm of reaction rate versus 1/temperature and then fit to a straight line, rather than trying to fit an exponential function directly.
In the specific question you mention, it is hard to know precisely what relationship is expected, but assuming the variable "t" is time, then t versus temperature could represent a cooling curve, in which case one might expect Newtonian cooling (i.e. temperature decreases at a rate proportional to difference in temperature of system and surroundings). This would lead to an exponential-type function, so plotting 1/time against temperature would give a curved line.