The paper compares direct and indirect model reference adaptive controllers (MRAC) with the L 1 adaptive controller. The architectures of the closed-loop systems are first compared in a simple setting that clarifies the similarities and differences of the controllers. The indirect MRAC and the L 1 controller have identical state predictors, but they differ in the computation of the control signal, which in L 1 is carried out by solving a feedback loop. The special case, where the controllers only adapt to a parameter representing input disturbances, is discussed. In this case, the closed-loop system is linear, so it may not be appropriate to call the controllers adaptive. The special case does, however, give good insight into similarities and differences of the controllers and the effects of various modifications. In particular, the analysis gives good understanding of the robustness properties.where x.t/ 2 R is the state of the system, u.t / 2 R is the control input, and a 2 R and b 2 R are unknown constants with known (conservative) bounds:jaj 6 a max , 0 < b min 6 b 6 b max .( 2 ) norm of a stable proper transfer function H.s/ with its impulse response h.t/ is defined as kH.s/k L1 D R 1 0 jh. /jd . || For a uniformly bounded signal a.t/ 2 R for t > 0, its L 1 norm is defined as kak L1 D sup t>0 ja.t/j. ** We have used the fact that for a stable proper transfer function H.s/ and a uniformly bounded input u.t/ 2 R, the output y.s/ D H.s/u.s/ can be uniformly bounded as follows: kyk L1 6 kH.s/k L1 kuk L1 .Bode plots of this transfer function for different adaptation gains are shown in Figure 4a. It is obvious that the Nyquist plot for L u 1 u 2 .s/ in Figure 4b never crosses the negative part of the real line; therefore, the closed-loop system has infinite gain margin (g m D 1). The gain crossover frequency ! gc can be computed from