Assuming universal ambiguity aversion, an extensive theoretical literature studies how ambiguity can account for market anomalies from the perspective of expected utility-based theories. We provide a systematic experimental assessment of ambiguity attitudes in different likelihood ranges, and in the gain domain, the loss domain and with mixed outcomes. We draw on a unified framework to elicit preferences across these domains. We replicate the usual finding of ambiguity aversion for moderate likelihood gains. However, when introducing losses or lower likelihoods, we observe predominantly ambiguity neutrality or seeking, rejecting universal ambiguity aversion.