What is the difference between merge, join, and concat in pandas?

How to think about it

What the interviewer wants to know

All three combine DataFrames, but they answer different questions. concat asks “stack these top-to-bottom (or side-by-side)”. merge asks “align rows where this column’s values match”. join is a convenience alias for merge when you want to align on the index. A strong answer names each one’s sweet spot and its failure mode.

The three tools, in plain English

concat — stack without matching. No key alignment. pandas lines up columns by name and fills gaps with NaN:

pd.concat([df1, df2])               # row-wise: stack vertically (axis=0)
pd.concat([df1, df2], axis=1)       # column-wise: side by side
pd.concat([df1, df2], ignore_index=True)  # reset to 0, 1, 2, ...

Use case: appending monthly export files, or adding a new set of features alongside existing ones.

merge — SQL-style join on any column(s). The most powerful and explicit option:

orders.merge(customers, on="customer_id", how="left")

# Different column names on each side
orders.merge(customers, left_on="cust_id", right_on="id")

# Multi-key join
df1.merge(df2, on=["year", "region"])

join — shorthand for index-on-index (or index-on-column) merge. Cleaner syntax when you have already set meaningful indexes:

df1.join(df2, how="left")   # aligns on index of both

# Join on a specific column of the right frame
df1.join(df2.set_index("customer_id"), on="cust_id")

join is implemented as a thin wrapper around merge — there is no performance difference.

Decision table

Scenario	Tool
Append rows from multiple DataFrames	concat(axis=0)
Add columns side-by-side	concat(axis=1)
Join on arbitrary column(s)	merge
Join on index	join or merge(left_index=True, right_index=True)
Join where column names differ	merge(left_on=..., right_on=...)

Playground — see all three in action

The key insight

The choice is almost always between concat (when you want to stack) and merge (when you want to align on keys). join is merge with a prettier syntax — use it when both DataFrames already have the right index set.

concat is deceptively powerful: it handles schema mismatches gracefully (columns not in both DataFrames get NaN), but that same flexibility can hide problems — a misnamed column in one batch will create two separate NaN-filled columns instead of merging cleanly.