There is a ksdensity function that can produce a kernel-smooth density estimate. The issue is that it produces a density (integrates to 1) and the histogram is not a density (bar heights sum to 1). You could figure out the area of the histogram and re-scale the ksdensity values. Alternatively, here's a way to create the histgram, normal curve, and kernel density separately:
x = [randn(100,1); 4+randn(50,1)];
[hts,ctrs] = hist(x)
bar(ctrs,hts,'hist')
area = sum(hts) * (ctrs(2)-ctrs(1))
xx = linspace(-3,7);
hold on; plot(xx,area*normpdf(xx,mean(x),std(x)),'r-')
f = ksdensity(x,xx);
plot(xx,area*f,'g-')
hold off
Best Answer