Flat rectangular tabs (n = 270) prepared from spherical (Tytin), admixed (Dispersalloy) or lathe-cut amalgam (ANA 2000) were subjected to aluminum oxide sandblasting with either 50-mu or 90-mu abrasive powder. Mandibular incisor edgewise brackets were bonded to these tabs. An intermediate resin was used, either All-Bond 2 Primers A + B or a 4-META product--Amalgambond Plus (AP) or Reliance Metal Primer (RMP)--followed by Concise. All specimens were stored in water at 37 degrees C for 24 hours and thermocycled 1000 times from 5 degrees C to 55 degrees C and back before tensile bond strength testing. The bond strength of Concise to etched enamel of extracted, caries-free premolars was used as a control. Bond failure sites were classified using a modified adhesive remnant index (ARI) system. Results were expressed as mean bond strength with SD, and as a function relating the probability of bond failure to stress by means of Weibull analysis. Mean tensile bond strength in the experimental groups ranged from 2.9 to 11.0 MPa--significantly weaker than the control sample (16.0 MPa). Bond failure invariably occurred at the amalgam/adhesive interface. The strongest bonds were created to the spherical and lathe-cut amalgams (range 6.8 to 11.0 MPa). Bonds to the spherical amalgam were probably more reliable. The intermediate application of the 4-META resins AP and RMP generally created significantly stronger bonds to all three basic types of amalgam products than the bonds obtained with the All-Bond 2 primers. The effect of abrasive-particle size on bond strength to different amalgam surfaces was not usually significant (p > 0.05). The implications of these findings are discussed in relationship to clinical experience bonding orthodontic attachments to large amalgam restorations in posterior teeth.