InterviewsVector

Pandas: drop vs Boolean Filtering vs df.filter() — What Each One Actually Does

Python

Quick answer

df.drop removes rows or columns by label — df.drop(index=[0,1]) or df.drop(columns=['a']). To keep rows by a CONDITION, use boolean masking: df[df['age'] > 30] (or df.query('age > 30')). df.filter() is a different tool: it selects labels by name with items=, like=, or regex= — it does NOT accept a boolean mask, so df.filter(df.a > 2) is wrong. Use drop to remove known labels, boolean masking to select by condition, and df.filter for name/pattern-based column/row selection.

Short answer: df.drop removes rows/columns by label. To keep rows by condition, use boolean masking (df[df.a > 2]), not df.filter. df.filter() is a third, different tool — it selects labels by name/regex (items=, like=, regex=) and does not accept a boolean mask.

These three get conflated constantly — including in tutorials that wrongly pass a boolean mask to df.filter(). Here's what each actually does.

drop — remove by label

import pandas as pd
df = pd.DataFrame({"A": [1, 2, 3, 4], "B": [5, 6, 7, 8], "C": [9, 10, 11, 12]})
 
df.drop(index=[1, 3])       # remove rows with labels 1 and 3
df.drop(columns=["B"])      # remove column B

drop operates on labels you name. df.drop([1, 3]) defaults to axis=0 (rows); use columns= (or axis=1) for columns.

Boolean masking — keep by condition

To keep rows that satisfy a condition, index with a boolean Series:

df[df["A"] > 2]             # keep rows where A > 2
df.loc[df["A"] > 2]        # same, explicit
df.query("A > 2")          # readable equivalent

This is what people usually mean by "filtering rows." Removing by condition is just the inverse mask (df[df["A"] <= 2]).

df.filter() — select labels by name/pattern

df.filter() is a different method: it selects rows or columns by their labels, not by a condition.

df.filter(items=["A", "C"])        # columns named A and C
df.filter(like="A", axis=1)        # columns whose name contains "A"
df.filter(regex="^A|C$", axis=1)   # columns matching a regex
df.filter(items=[0, 2], axis=0)    # rows with index labels 0 and 2
# ❌ WRONG — filter does not take a boolean mask
df.filter(df["A"] > 2)
 
# ✅ use boolean masking instead
df[df["A"] > 2]

Which to use

GoalUse
Remove specific known rows/columnsdf.drop(index=…) / df.drop(columns=…)
Keep rows meeting a conditiondf[mask] / df.query(...)
Select columns/rows by name or patterndf.filter(items=/like=/regex=)

Related guides

Sources

Key takeaways

  • •df.drop removes by LABEL: df.drop(index=[...]) for rows, df.drop(columns=[...]) for columns.
  • •To keep rows by a CONDITION, use boolean masking: df[df['a'] > 2] — not df.filter.
  • •df.filter() selects LABELS via items=/like=/regex=; it does NOT take a boolean mask.
  • •df.filter(df['a'] > 2) is a bug — passing a mask to filter doesn't do condition-based row selection.
  • •drop and boolean masking are opposites: drop names what to remove, masking names what to keep.

Frequently asked questions

What's the difference between drop and filtering in pandas?

df.drop removes rows or columns by their labels (df.drop(index=[1,3]) or df.drop(columns=['B'])). Boolean 'filtering' keeps rows that satisfy a condition (df[df['age'] > 30]). One names what to remove; the other names what to keep.

Does df.filter() filter rows by a condition?

No. df.filter() selects labels by name — items=['a','b'] for exact names, like='2023' for substrings, regex='e$' for patterns — along an axis. It does not take a boolean mask, so df.filter(df.a > 2) does not do condition-based row selection. Use df[df.a > 2] for that.

How do I remove rows by condition in pandas?

Keep the rows you want with a boolean mask — df = df[df['age'] >= 18] — which is the inverse of dropping. If you specifically want drop syntax, compute the labels to remove first: df.drop(index=df[df['age'] < 18].index).

By Mohammad Wasi

Software Engineering Leader & Technical Author · Updated August 26, 2026

See all posts

Related Posts