In general, the inverting configuration, with a resistor in series between the input and summing point, has inferior noise performance due to the voltage noise of a relatively large resistor, but does not suffer from common-mode-induced distortion.
The non-inverting configuration is therefore usually preferred for noise-sensitive applications, since the resistors in the feedback network can be made smaller than what one would use in the inverting configuration. The common-mode problem can be important, since the common-mode voltage equals the input signal voltage. If there be a non-linear input current due to the common-mode voltage (common with BJTs), it flows through the source impedance (including any input coupling capacitor). The voltage from that current looks exactly like an input voltage, and will not be affected by feedback.
An interesting variation, almost never used, is to place the input signal between the feedback voltage divider and the inverting input, with the non-inverting input grounded, thus avoiding both problems but requiring a non-grounded source (possibly a phono pickup, microphone, or transformer secondary).