What Is the Radial Arm Maze?
The radial arm maze (RAM) was introduced by Olton and Samuelson in 1976 and is one of the most widely used tasks for dissociating working memory from reference memory in rodents. The apparatus consists of a central platform (typically 30-40 cm diameter) with 8 arms radiating outward like spokes of a wheel, each arm approximately 50 cm long and 10 cm wide with walls to prevent the animal from jumping between arms.
In the standard protocol, a subset of arms (commonly 4 of 8) is consistently baited with food rewards across all training sessions, while the remaining arms are never baited. The animal is food-restricted to 80-85% of free-feeding weight and placed on the central platform at the start of each trial. It must navigate the arms to collect all food rewards as efficiently as possible.
Spatial cues (posters, objects, room features) placed around the testing room allow the animal to form allocentric spatial representations. Over 10-20 training days (typically 1 trial per day), animals with intact hippocampal function learn which arms are baited (reference memory) and avoid revisiting arms within a trial (working memory). The RAM is particularly valued for its ability to independently assess these two memory systems within a single task.
Working Memory vs. Reference Memory Errors
The radial arm maze uniquely allows independent measurement of two dissociable memory systems.
Working memory errors are re-entries into arms that have already been visited within the current trial. These errors reflect a failure of short-term, trial-specific memory. The animal must maintain an updated representation of which arms it has already checked during the ongoing session. Working memory errors should decrease within a session and are sensitive to prefrontal cortex and hippocampal manipulations.
Reference memory errors are entries into arms that are never baited across any training session. These errors reflect a failure of long-term, across-session memory. The animal must learn and retain which arms consistently contain food and which never do. Reference memory errors should decrease across training days and are primarily sensitive to hippocampal lesions.
A third category — repeated reference memory errors (re-entering a never-baited arm already visited in the current trial) — combines both error types and is sometimes reported separately. This tool counts entries to never-baited arms as reference memory errors regardless of whether they are first or repeated visits, which is the most common scoring convention (Olton 1979; Jarrard 1993).
Scoring Methodology
Scoring the radial arm maze requires recording the complete sequence of arm entries for each trial. An arm entry is typically defined as the animal placing all four paws into the arm.
Working memory errors (WME): Count the number of re-entries to baited arms that were already visited earlier in the same trial. For example, if arms 1,2,4,7 are baited and the sequence is 1,3,2,1,4,7 — the second visit to arm 1 is a working memory error.
Reference memory errors (RME): Count the number of entries to arms that are never baited. In the example above, arm 3 is never baited, so entering arm 3 is a reference memory error.
Percent correct first entries: Of the first N arm choices (where N = number of baited arms), how many were to baited arms? This metric captures initial choice accuracy without being confounded by later exploratory or perseverative behavior. It is analogous to the primary error concept in the Barnes maze.
Total errors: WME + RME. This provides an overall performance metric but does not distinguish between memory systems.
All metrics should be reported with group means and SEM across animals. Learning curves plotting errors across training days are essential for demonstrating acquisition. A minimum of 8-10 training days is recommended to observe reliable learning, with many protocols using 15-20 days.
