The supplied plot function that is fit -object aware uses an internally-determined number of points to evaluate the function that are more than those in the data set itself--I just did a test here--for a sample dataset of length 15 where I fit y=x.^2+rand(size(x))*10; for a vector x=[1:15];
>> f=fit(x',y','poly2'); % fit the above quad+noise
>> h=plot(f,x,y) % use the builtin plot; return line handles
h =
633.0012
634.0002
>> tmp=get(h(1),'xdata'); % get the data for the raw data...
>> length(tmp) % and show how many points were there--
ans =
15
>> tmp=get(h(2),'xdata'); % ditto for the curve of the fit...
>> length(tmp)
ans =
1013
>>
As you can see, it computed the fit at over 1000 points internally to draw a smooth curve whereas your manual plot is just "connecting the dots" between the fitted points since you only evaluated at the input values.
