Investors with a fixed budget must decide how much to put into each of several stocks, balancing return against risk. The classical answer, Markowitz's mean–variance model, is a quadratic programme that becomes heavy as the number of stocks grows. This paper reviews two published studies that solve the same problem with linear programming instead. Papahristodoulou compares a quadratic model with a maximin model and a mean-absolute-deviation (MAD) model on five Stockholm stocks. Kucukbay and Araz build a portfolio from 89 Borsa Istanbul stocks using linear physical programming (LPP) and fuzzy goal programming (FGP). For each study we set out the real-life problem, the mathematical model, the solution method, the results and the weaknesses. In the first study the author reports the MAD portfolio as close to the quadratic one while avoiding the covariance matrix, although on the published figures its risk sits nearer to that of maximin. In the second, the LPP portfolio had the best Sharpe index in the test year (0.43, against 0.36 for FGP and 0.26 for the market index). Both studies rest on small samples and short test periods, so their results show that linear models are practical, not that they are best.