This paper discusses the results of an extensive experimental study of the high-frequency characteristics of several commercial (GaAl)As injection lasers. The lasers tested included the Hitachi buried heterostructure (BH), the Hitachi channel substrate planar (CSP), the Mitsubishi transverse junction stripe (TJS), the General Optronics stripe (GO), and the RCA constricted double heterostructure (CDH). We find that the maximum practical analog modulation frequency f o is in the range of 5 GHz. This practical modulation frequency is limited by a combination of laser-relaxation resonance, laser parasitics, and drive current necessary for long-life operation. The relationship between laser-relaxation oscillation resonance frequency f o and drive current can be expressed as f_{o} = K(I/Ith - 1)^{1/2} GHz where I and Ith are the laser drive and threshold currents, respectively. The value of K is experimentally determined to lie in the range 3.0 K j and a capacitance C j . The product of R_{j}C_{j} has been experimentally determined for both the Hitachi CSP and BH laser and lies in the range of 1-1.1 \times 10^{-10} s. The effect of laser mount and packaging can also be included in the equivalent circuit by the addition of series inductors and shunt capacitors. Lastly, all injection lasers studied displayed a dip in the small signal modulation response prior to the onset of the relaxation oscillation resonance. In particular, there is a pronounced dip even for lasers with good lateral carrier confinement at dc, such as the buried heretostructure (BH). Previous workers have attributed the dip in the modulation response as resulting from either lateral diffusion of injected carriers, or the effect of junction capacitance. Our experimental results indicate that while the effect of the junction capacitance may play some role in explaining the dip for the BH laser, it is not the dominant mechanism. We speculate that lasers with good lateral carrier confinement at dc (BH) may suffer a significant degradation of carrier confinement at gigahertz rates. A simple qualitative model is proposed to explain this mechanism and methods for reducing the effect are discussed.