Metal magnetic memory (MMM) testing method has been proved to be a valid approach to monitor early damage and predict fatigue life, but there is no systematic description and comparison about the characteristics of MMM signals for welding and non-welding specimens with different materials under tension-compression fatigue. Thus, the fatigue tests for Q345B and Q345qC welding and non-welding specimens were carried out and the MMM testing was done synchronously. Then, based on the normal components of MMM signals, H
p
(y), and its variations, ΔH
p
(y), the slope of ΔH
p
(y) fitting curves, K
C
, and the average of ΔH
p
(y) signals on the whole testing line, ΔH
p
(y)
ave
, were extracted as the characteristic parameters for Q345B and Q345qC non-welding specimens, respectively. Additionally, the gradient of “peak–vale” on H
p
(y) signals curves, k, and the average of H
p
(y) signals in the scope of welding zone, H
p
(y)
ave
, were extracted as the characteristic parameters for Q345B and Q345qC welding specimens, respectively. Some conclusions can be drawn by comparison: the variation laws of H
p
(y) signals with cyclic loads are related to the material. The characteristic parameters for welding and non-welding specimens with the same material have a certain similarity. Then, the failure of non-welding steel can be warned early by K
C
or ΔH
p
(y)
ave
when the fatigue life has 20% left, but the damage degree of welding specimen should begin to be valued highly once the k or H
p
(y)
ave
is decreasing. Finally, the fatigue life of welding and non-welding specimens with different materials can be evaluated effectively by using corresponding magnetic characteristic parameters.
